Built-in Plugins

The platform ships with 105 plugins under packages/plugins/, spanning 21 of the 27 categories declared by PLUGIN_CATEGORIES in packages/plugin/src/contracts/plugin-manifest.types.ts: AI provider, search, content extractor, screenshot, git provider, deployment, data source, pipeline, storage, database, vector store, DNS, secret store resolver, job runtime, email provider, notification channel, connector, metrics, utility, build, and identity. (form, integration, theme, memory, rag, and app-dependency are declared as contracts but have no shipped plugin yet.) This page documents the most widely used ones, with configuration and environment variables for each.

Plugin count by category

Counted from each package's everworks.plugin.category field.

Category	Count	Plugin IDs
pipeline	14	activepieces, agent-pipeline, claude-code, claude-managed-agent, codex, composio, gemini, gtm-pipeline, hermes-agent, make, opencode, sim-ai, standard-pipeline, zapier
ai-provider	11	anthropic, google, grok, groq, lm-studio, mistral, ollama, openai, openrouter, vercel-ai-gateway, vllm
connector	11	bluesky-connector, discord-connector, google-workspace-connector, hubspot-connector, jira-connector, linear-connector, mastodon-connector, notion-connector, pipedrive-connector, slack-connector, zoom-connector
utility	11	agentmemory, browser-automation, comparison-generator, everworks-playbooks, everworks-skills, everworks-task-tracker, langfuse, local-workspace, memory-pipeline-modifier, pty-local, sandbox-workspace
search	9	brave, brightdata, exa, firecrawl, linkup, perplexity, serpapi, tavily, valyu
secret-store-resolver	7	secret-store-aws-sm, secret-store-azure-kv, secret-store-doppler, secret-store-gcp-sm, secret-store-infisical, secret-store-k8s, secret-store-vault
job-runtime	6	job-runtime-bullmq, job-runtime-inngest, job-runtime-node, job-runtime-pgboss, job-runtime-temporal, job-runtime-trigger
content-extractor	6	jina, local-content-extractor, notion-extractor, officecli-extractor, pdf-extractor, scrapfly
notification-channel	5	discord-channel, novu-channel, slack-channel, telegram-channel, whatsapp-channel
email-provider	5	mailchimp-transactional, mailgun, postmark, resend, sendgrid
storage	4	aws-s3, github-storage, local-fs, minio
metrics	4	custom-http-metrics, google-analytics-metrics, posthog-metrics, stripe-metrics
vector-store	2	pgvector, qdrant
screenshot	2	screenshotone, urlbox
deployment	2	k8s, vercel
git-provider	1	github
database	1	postgres-db
data-source	1	apify
dns	1	cloudflare-dns
build	1	github-actions-build
identity	1	oidc-identity

A plugin's capabilities are independent of its category: scrapfly sits in content-extractor but also advertises screenshot, and every search plugin except brave, perplexity and serpapi also advertises content-extractor.

EW-693 — Core vs distributable

As of EW-693 (Dynamic Plugin Distribution), every plugin is classified as either core (always bundled in the platform image, present in both bundled and dynamic modes) or distributable (published to the npm registry under @ever-works/*, installable at runtime when PLUGIN_DISTRIBUTION_MODE=dynamic). The classification lives on the plugin's manifest (everworks.plugin.distribution); when absent, the default is core for systemPlugin: true and registry otherwise.

Core (27): agent-pipeline, comparison-generator, github, job-runtime-bullmq, job-runtime-inngest, job-runtime-node, job-runtime-pgboss, job-runtime-temporal, job-runtime-trigger, k8s, local-content-extractor, local-fs, local-workspace, openrouter, pgvector, postgres-db, sandbox-workspace, secret-store-aws-sm, secret-store-azure-kv, secret-store-doppler, secret-store-gcp-sm, secret-store-infisical, secret-store-k8s, secret-store-vault, standard-pipeline, tavily, vercel. These are bundled in every image. local-fs is the default boot-storage so the API can serve without any distributable storage plugin enabled (FR-4).

Distributable (78): every other plugin under packages/plugins/*. In bundled mode these still ship in the image (so a fresh deploy behaves byte-for-byte the same as pre-EW-693); in dynamic mode they are stripped from the image and pulled from @ever-works/<id>-plugin on first enable via the per-replica installer. The full split is generated from each plugin's package.json everworks.plugin manifest — see scripts/strip-non-core-plugins.js for the runtime classification rule.

27 core plus 78 distributable accounts for all 105 plugins. Sixteen plugins state distribution: 'core' explicitly (the six job runtimes, the seven secret stores, local-fs, pgvector, postgres-db); the other eleven inherit core from systemPlugin: true. Everything added since — including hermes-agent, gtm-pipeline, officecli-extractor, grok, qdrant, cloudflare-dns, the connectors, the email providers, the notification channels and the metrics providers — is registry, so it is stripped from a dynamic-mode image and installed on demand.

See docs/specs/features/dynamic-plugin-distribution/spec.md for the full feature spec and docs/internal/EW-693-deployment.md for the operator runbook.

AI Providers

AI provider plugins implement ai-provider capability and power all content generation, chat, and structured output features.

OpenAI

Use OpenAI models (GPT-5.1, GPT-5-nano, GPT-4o-mini) for content generation and AI features.

Field	Value
Plugin ID	openai
Configuration Mode	user-required
Auto Enable	No

Settings:

Setting	Type	Default	Description
apiKey	string	—	OpenAI API key (required, secret)
defaultModel	string	gpt-5.1	Default model for all tasks
simpleModel	string	gpt-5-nano	Model for tags, short descriptions
mediumModel	string	gpt-4o-mini	Model for summaries, reformatting
complexModel	string	gpt-5.1	Model for full page generation
temperature	number	0.7	Response variability (0–2)
maxTokens	number	4096	Max response length
baseUrl	string	https://api.openai.com/v1	API endpoint

Anthropic

Use Anthropic Claude models for content generation.

Field	Value
Plugin ID	anthropic
Configuration Mode	user-required
Auto Enable	No

Settings:

Setting	Type	Default	Description
apiKey	string	—	Anthropic API key (required, secret)
defaultModel	string	claude-sonnet-4-5-20250514	Default model
simpleModel	string	claude-haiku-4-5-20251001	Simple tasks model
mediumModel	string	claude-sonnet-4-5-20250929	Standard tasks model
complexModel	string	claude-sonnet-4-5-20250514	Complex tasks model
temperature	number	0.7	Response variability
maxTokens	number	4096	Max response length
baseUrl	string	https://api.anthropic.com/v1/	API endpoint

Google Gemini

Use Google Gemini models for content generation.

Field	Value
Plugin ID	google
Configuration Mode	user-required
Auto Enable	No

Settings:

Setting	Type	Default	Description
apiKey	string	—	Google API key (required, secret)
defaultModel	string	models/gemini-2.5-flash	Default model
simpleModel	string	models/gemini-2.0-flash	Simple tasks model
mediumModel	string	models/gemini-2.5-flash	Standard tasks model
complexModel	string	models/gemini-2.5-pro	Complex tasks model
temperature	number	0.7	Response variability
maxTokens	number	4096	Max response length
baseUrl	string	https://generativelanguage.googleapis.com/v1beta/openai/	API endpoint

Groq

Use Groq for fast AI inference with open-source models.

Field	Value
Plugin ID	groq
Configuration Mode	user-required
Auto Enable	No

Settings:

Setting	Type	Default	Description
apiKey	string	—	Groq API key (required, secret)
defaultModel	string	meta-llama/llama-4-scout-17b-16e-instruct	Default model
baseUrl	string	https://api.groq.com/openai/v1	API endpoint

Grok (xAI)

Use xAI's Grok models through the OpenAI-compatible xAI API — 131,072-token context, vision input, tool calling and structured output. Assign a different Grok model per task tier.

Field	Value
Plugin ID	grok
Category	ai-provider
Configuration Mode	user-required
Auto Enable	No
Distribution	registry

Settings:

Setting	Type	Default	Description
apiKey	string	—	xAI API key (required, secret, user-scoped; env fallback XAI_API_KEY)
defaultModel	string	grok-2-latest	Default model for all AI tasks (required)
simpleModel	string	grok-2-latest	Tags, short descriptions, quick classifications
mediumModel	string	grok-2-latest	Listings, summaries, content reformatting
complexModel	string	grok-2-latest	Full page generation and multi-step analysis
baseUrl	string	https://api.x.ai/v1	API endpoint (advanced, hidden by default)
temperature	number	0.7	Response variability (0–2, hidden)
maxTokens	number	4096	Max response length (hidden)

Get a key at console.x.ai, then enable the plugin at Settings → Plugins → AI Provider (/settings/plugins/ai-provider) and paste it into xAI API Key.

Ollama

Use locally running models via Ollama. No API key required.

Field	Value
Plugin ID	ollama
Configuration Mode	user-required
Auto Enable	No

Settings:

Setting	Type	Default	Description
baseUrl	string	—	Ollama URL (required, e.g., http://localhost:11434/v1)
defaultModel	string	llama2	Default model
apiKey	string	ollama	API key (optional, defaults to ollama)

LM Studio

Use locally running models via the LM Studio local server. No API key required.

Field	Value
Plugin ID	lm-studio
Configuration Mode	user-required
Auto Enable	No

Settings:

Setting	Type	Default	Description
baseUrl	string	—	LM Studio URL (required, e.g., http://localhost:1234/v1)
apiKey	string	lm-studio	API key (optional; only for an auth proxy in front of LM Studio)
defaultModel	string	—	Loaded model id (required; populated via the model picker)

vLLM

Use a self-hosted vLLM OpenAI-compatible server (typically GPU-hosted).

Field	Value
Plugin ID	vllm
Configuration Mode	user-required
Auto Enable	No

Settings:

Setting	Type	Default	Description
baseUrl	string	—	vLLM URL (required, e.g., http://localhost:8000/v1)
apiKey	string	EMPTY	API key (secret; only required if started with --api-key)
defaultModel	string	—	Model id passed to vllm serve --model (required; via model picker)

Mistral

Use Mistral AI models for content generation.

Field	Value
Plugin ID	mistral
Configuration Mode	user-required
Auto Enable	No

Settings:

Setting	Type	Default	Description
apiKey	string	—	Mistral API key (required, secret)
defaultModel	string	mistral-small-latest	Default model for all tasks
simpleModel	string	mistral-small-latest	Model for tags, short descriptions
mediumModel	string	mistral-medium-latest	Model for summaries, reformatting
complexModel	string	mistral-large-latest	Model for full page generation
temperature	number	0.7	Response variability (0–2)
maxTokens	number	4096	Max response length
baseUrl	string	https://api.mistral.ai/v1	API endpoint

AI Gateways

AI gateway plugins route requests through multi-provider services, giving access to many models through a single API key.

OpenRouter

Access 400+ models from multiple providers through OpenRouter.

Field	Value
Plugin ID	openrouter
Configuration Mode	hybrid
Auto Enable	Yes
Default For	ai-provider
System Plugin	Yes

Environment Variables:

Variable	Required	Description
PLUGIN_OPENROUTER_API_KEY	Yes	OpenRouter API key
PLUGIN_OPENROUTER_DEFAULT_MODEL	No	Override default model
PLUGIN_OPENROUTER_SIMPLE_MODEL	No	Override simple tasks model
PLUGIN_OPENROUTER_MEDIUM_MODEL	No	Override medium tasks model
PLUGIN_OPENROUTER_COMPLEX_MODEL	No	Override complex tasks model
PLUGIN_OPENROUTER_BASE_URL	No	API endpoint (default: https://openrouter.ai/api/v1)

Settings:

Setting	Type	Default	Description
apiKey	string	—	OpenRouter API key (secret)
defaultModel	string	openai/gpt-5.1	Default model
simpleModel	string	openai/gpt-5-nano	Simple tasks model
mediumModel	string	openai/gpt-4o	Standard tasks model
complexModel	string	openai/gpt-5.1	Complex tasks model

Vercel AI Gateway

Route AI requests through Vercel's AI Gateway.

Field	Value
Plugin ID	vercel-ai-gateway
Configuration Mode	hybrid
Auto Enable	No

Environment Variables:

Variable	Required	Description
PLUGIN_VERCEL_AI_GATEWAY_API_KEY	Yes	API key
PLUGIN_VERCEL_AI_GATEWAY_BASE_URL	No	Endpoint (default: https://ai-gateway.vercel.sh/v1)

Search

Search plugins power web research during the generation pipeline.

Tavily

Web search and content extraction optimized for AI applications. This is the default search provider.

Field	Value
Plugin ID	tavily
Configuration Mode	hybrid
Auto Enable	Yes
Default For	search
System Plugin	Yes
Capabilities	search, content-extractor

Environment Variables:

Variable	Required	Description
PLUGIN_TAVILY_API_KEY	No	Tavily API key (can be set in user settings instead)

Settings:

Setting	Type	Default	Description
apiKey	string	—	Tavily API key (required, secret)

Brave Search

Web search using the Brave Search API.

Field	Value
Plugin ID	brave
Configuration Mode	hybrid
Auto Enable	No

Environment Variables:

Variable	Required	Description
PLUGIN_BRAVE_API_KEY	No	Brave Search API key

Settings:

Setting	Type	Default	Description
apiKey	string	—	Brave API key (required, secret)
maxResults	number	10	Results per search (1–20)

SerpAPI

Web search using SerpAPI with support for multiple search engines.

Field	Value
Plugin ID	serpapi
Configuration Mode	hybrid
Auto Enable	No

Environment Variables:

Variable	Required	Description
PLUGIN_SERPAPI_API_KEY	No	SerpAPI key

Settings:

Setting	Type	Default	Description
apiKey	string	—	SerpAPI key (required, secret)
engine	string	google	Search engine: google, bing, yahoo, duckduckgo, baidu, yandex
maxResults	number	10	Results per search (1–100)

Exa

AI-native search with neural and keyword modes.

Field	Value
Plugin ID	exa
Configuration Mode	hybrid
Auto Enable	No
Capabilities	search, content-extractor

Environment Variables:

Variable	Required	Description
PLUGIN_EXA_API_KEY	No	Exa API key

Settings:

Setting	Type	Default	Description
apiKey	string	—	Exa API key (required, secret)
searchType	string	auto	Search type: auto, neural, keyword
maxResults	number	10	Results per search (1–100)
category	string	—	Filter by category: company, research paper, news, tweet, personal site, github

Perplexity

AI-powered web search with citations via the Perplexity API.

Field	Value
Plugin ID	perplexity
Configuration Mode	hybrid
Auto Enable	No

Environment Variables:

Variable	Required	Description
PLUGIN_PERPLEXITY_API_KEY	No	Perplexity API key

Settings:

Setting	Type	Default	Description
apiKey	string	—	Perplexity API key (required, secret)

Bright Data

Web search and content extraction via the Bright Data SERP API and Web Scraper.

Field	Value
Plugin ID	brightdata
Configuration Mode	hybrid
Auto Enable	No
Capabilities	search, content-extractor

Environment Variables:

Variable	Required	Description
PLUGIN_BRIGHTDATA_API_KEY	No	Bright Data API key

Settings:

Setting	Type	Default	Description
apiKey	string	—	Bright Data API key (required, secret)

Firecrawl

Web search and markdown content extraction via the Firecrawl API.

Field	Value
Plugin ID	firecrawl
Configuration Mode	hybrid
Auto Enable	No
Capabilities	search, content-extractor

Environment Variables:

Variable	Required	Description
PLUGIN_FIRECRAWL_API_KEY	No	Firecrawl API key

Settings:

Setting	Type	Default	Description
apiKey	string	—	Firecrawl API key (required, secret)

Valyu

AI-native multi-source search and content extraction via the Valyu API.

Field	Value
Plugin ID	valyu
Configuration Mode	hybrid
Auto Enable	No
Capabilities	search, content-extractor

Environment Variables:

Variable	Required	Description
PLUGIN_VALYU_API_KEY	No	Valyu API key

Settings:

Setting	Type	Default	Description
apiKey	string	—	Valyu API key (required, secret)
responseLength	string	medium	Content volume per result: short, medium, large, max

Linkup

Web search and content extraction via the Linkup API. Optimized for AI-precision results and clean content extraction from any URL.

Field	Value
Plugin ID	linkup
Configuration Mode	hybrid
Auto Enable	No
Capabilities	search, content-extractor

See Linkup Plugin for setup. Refer to packages/plugins/linkup/src/ for the current settings schema.

Git Provider

GitHub

Repository management, cloning, pushing, pull requests, and OAuth authentication. This is the default git provider and is always enabled.

Field	Value
Plugin ID	github
Configuration Mode	admin-only
Auto Enable	Yes
Default For	git-provider
System Plugin	Yes
Capabilities	git-provider, oauth

Environment Variables:

Variable	Required	Description
PLUGIN_GITHUB_CLIENT_ID	No	GitHub OAuth App client ID
PLUGIN_GITHUB_CLIENT_SECRET	No	GitHub OAuth App client secret

Settings:

Setting	Type	Default	Description
clientId	string	—	GitHub OAuth client ID
clientSecret	string	—	GitHub OAuth client secret (secret)
apiBaseUrl	string	https://api.github.com	GitHub API endpoint

Deployment

Vercel

Deploy work websites to Vercel. This is the default deployment provider and is always enabled.

Field	Value
Plugin ID	vercel
Configuration Mode	user-required
Auto Enable	Yes
Default For	deployment
System Plugin	Yes

Settings:

Setting	Type	Default	Description
apiToken	string	—	Vercel API token (required, secret)
defaultTeamScope	string	—	Default Vercel team scope (optional)

Kubernetes

Deploy work websites to any Kubernetes cluster you control as an alternative to Vercel. Selectable per-work via deployProvider: k8s (dashboard or .works/works.yml). Supports a pluggable container registry (GitHub Container Registry by default) and ingress controller strategies (nginx, Traefik, plus a generic fallback). See Kubernetes Deployment for the full user guide.

Field	Value
Plugin ID	k8s
Configuration Mode	user-required
Auto Enable	Yes
Default For	—
System Plugin	Yes

Settings:

Setting	Type	Default	Description
kubeconfig	string	—	Full kubeconfig YAML (required, secret, user-scoped)
kubeContext	string	—	Override the kubeconfig's current-context
namespace	string	ever-works	Target Kubernetes namespace
registry.kind	string	github	One of github, dockerhub, generic
registry.owner	string	(auto)	GHCR owner; defaults to your connected GitHub account
registry.visibility	string	auto	auto mirrors the website repo, or public / private
ingressClass	string	(cluster default)	Detected at validation time; populated from IngressClass list
ingressHost	string	—	Default ingress host when a work has no custom domain
tlsIssuer	string	—	cert-manager ClusterIssuer name
replicas	integer	1	Pod replicas (1–10)

Screenshot

Screenshot plugins capture website images for work item previews.

ScreenshotOne

Website screenshots via the ScreenshotOne API.

Field	Value
Plugin ID	screenshotone
Configuration Mode	hybrid
Auto Enable	No

Environment Variables:

Variable	Required	Description
PLUGIN_SCREENSHOTONE_ACCESS_KEY	No	ScreenshotOne access key
PLUGIN_SCREENSHOTONE_SECRET_KEY	No	ScreenshotOne secret key

Settings:

Setting	Type	Default	Description
accessKey	string	—	Access key (required, secret)
secretKey	string	—	Secret key (secret)
viewportWidth	number	1280	Viewport width
viewportHeight	number	1024	Viewport height
format	string	png	Image format
blockAds	boolean	true	Block ads
blockTrackers	boolean	true	Block trackers

URLBox

Website screenshots via the URLBox API.

Field	Value
Plugin ID	urlbox
Configuration Mode	hybrid
Auto Enable	No

Environment Variables:

Variable	Required	Description
PLUGIN_URLBOX_API_KEY	No	URLBox API key
PLUGIN_URLBOX_API_SECRET	No	URLBox API secret

Settings:

Setting	Type	Default	Description
apiKey	string	—	URLBox API key (required, secret)
apiSecret	string	—	URLBox API secret (secret)
viewportWidth	number	1280	Viewport width (320–3840)
viewportHeight	number	1024	Viewport height (200–2160)
format	string	png	Image format
blockAds	boolean	true	Block ads
hideCookieBanners	boolean	true	Hide cookie consent banners

Content Extractors

Content extractor plugins fetch and parse web page content for the generation pipeline.

Local Content Extractor

Built-in HTML content extraction using fetch and HTML parsing. No external API needed. This is the default content extractor and is always enabled.

Field	Value
Plugin ID	local-content-extractor
Configuration Mode	admin-only
Auto Enable	Yes
Default For	content-extractor
System Plugin	Yes

Settings:

Setting	Type	Default	Description
timeout	number	15000	Request timeout in ms (1000–60000)
minContentLength	number	200	Minimum content length (0–10000)
userAgent	string	—	Custom user agent string

Notion Extractor

Extract content from Notion pages (both public and private).

Field	Value
Plugin ID	notion-extractor
Configuration Mode	hybrid
Auto Enable	No

Settings:

Setting	Type	Default	Description
apiKey	string	—	Notion API key (optional, secret — needed for private pages)
useSplitbeeForPublicPages	boolean	true	Use Splitbee API for public pages
timeout	number	15000	Request timeout in ms

Jina AI

Web search and content extraction via Jina AI's reader and search APIs.

Field	Value
Plugin ID	jina
Configuration Mode	hybrid
Auto Enable	No
Capabilities	search, content-extractor

Environment Variables:

Variable	Required	Description
PLUGIN_JINA_API_KEY	No	Jina API key

Settings:

Setting	Type	Default	Description
apiKey	string	—	Jina API key (required, secret)

Scrapfly

Website screenshot capture and content extraction via the Scrapfly API.

Field	Value
Plugin ID	scrapfly
Configuration Mode	hybrid
Auto Enable	No
Capabilities	screenshot, content-extractor

Environment Variables:

Variable	Required	Description
PLUGIN_SCRAPFLY_API_KEY	No	Scrapfly API key

Settings:

Setting	Type	Default	Description
apiKey	string	—	Scrapfly API key (required, secret)

PDF Content Extractor

Extract text content from PDF files. Uses text-layer extraction by default, with optional OCR fallback via Mistral AI for scanned or image-based PDFs.

Field	Value
Plugin ID	pdf-extractor
Auto Enable	No
Capabilities	content-extractor

Environment Variables:

Variable	Required	Description
PLUGIN_PDF_EXTRACTOR_API_KEY	No	Mistral AI API key (only needed for OCR fallback)

Settings:

Setting	Type	Default	Description
mistralApiKey	string	—	Mistral API key for OCR fallback (optional, secret)

OfficeCLI Extractor

Extract text from Office documents — Word .docx, Excel .xlsx, PowerPoint .pptx — via the OfficeCLI tool and its official @officecli/sdk Node SDK. Off by default; enable it only when you need Office source material. It has zero URL overlap with the PDF extractor, and OfficeCLI ships musl binaries so it runs on the platform's node:22-alpine base image.

Field	Value
Plugin ID	officecli-extractor
Category	content-extractor
Auto Enable	No
System Plugin	No
Capabilities	content-extractor

Settings:

Setting	Type	Default	Description
renderMode	string	text	Output format: text or markdown
maxBytes	number	26214400	Max document size to download and process, 25 MB (env OFFICECLI_EXTRACTOR_MAX_BYTES)
timeout	number	30000	HTTP download + OfficeCLI command timeout in ms (5000–300000, hidden)
binaryPath	string	—	Absolute path to a specific officecli binary; blank uses the bundled one (hidden)

When a source URL points at an Office document, the content-extractor facade delegates to this plugin: it downloads the file behind an SSRF guard and the byte cap, writes it to a private temp file, opens it with OfficeCLI, and always cleans up the temp file and the OfficeCLI resident afterwards.

The bundled OfficeCLI binary and SDK are Apache-2.0; the Ever Works wrapper is AGPL-3.0. See packages/plugins/officecli-extractor/README.md for the full attribution notice.

Data Source

Apify

Import data from external sources using Apify web scraping actors.

Field	Value
Plugin ID	apify
Capabilities	data-source, form-schema-provider

Settings:

Setting	Type	Default	Description
apiToken	string	—	Apify API token (required, secret)
defaultFieldMapping	object	—	Field mapping (name, description, source_url, category, image_url)

Storage

Storage plugins implement the storage capability plus the object verbs (put-object, get-object, presigned-put where the backend supports pre-signed uploads, and put-object-stream / get-object-stream where the backend can write and read an object as a stream rather than as one buffer). They back every file the dashboard accepts — Knowledge Base documents, item images, avatars. Pick one at Settings → Plugins → Storage (/settings/plugins/storage). Introduced in EW-637; the Git LFS and data-repo work landed in EW-644; the two streaming verbs were added by the workspace-backup archive (AW-22), which can run to gigabytes and so may never pass through memory. Both are optional: a consumer probes for the method rather than checking a backend id, and a backend without them simply carries a smaller size ceiling. See Storage Backends for the user-facing guide.

Local Filesystem

Writes objects to a directory on the API server. This is the default boot storage (FR-4): the API can serve with no distributable storage plugin enabled at all.

Field	Value
Plugin ID	local-fs
Category	storage
Auto Enable	Yes
System Plugin	Yes
Distribution	core
Capabilities	storage, put-object, get-object, put-object-stream, get-object-stream

Settings:

Setting	Type	Default	Environment variable	Description
uploadsDir	string	<tmpdir>/ever-works-uploads	UPLOADS_DIR	Absolute path on the API server for objects
maxBytes	number	5242880 (5 MiB)	UPLOADS_MAX_BYTES	Per-object size cap in bytes

Streaming writes go to a temporary name inside the same owner directory and are renamed into place, so the on-disk layout (<UPLOADS_DIR>/<ownerId>/<sha256>.<ext>) is identical whichever verb wrote the object.

Single-node only — the directory is local to the API pod, so a multi-replica deployment needs S3, MinIO, or GitHub storage instead.

AWS S3

Field	Value
Plugin ID	aws-s3
Category	storage
Auto Enable	No
Distribution	registry
Capabilities	storage, put-object, get-object, presigned-put

Settings (required: region, bucket):

Setting	Type	Default	Environment variable	Description
region	string	—	AWS_S3_REGION	Region of the bucket, e.g. us-east-1
bucket	string	—	AWS_S3_BUCKET	Bucket name
accessKeyId	string	—	AWS_ACCESS_KEY_ID	IAM access key (secret); omit to use the AWS credential chain
secretAccessKey	string	—	AWS_SECRET_ACCESS_KEY	IAM secret (secret); omit to use the AWS credential chain
presignExpiresSeconds	number	600	AWS_S3_PRESIGN_EXPIRES_SECONDS	Pre-signed upload URL TTL, 60–3600

Leaving both key fields blank is the recommended production setup: the plugin then falls back to the default AWS credential chain (instance role, IRSA, or the ambient profile).

MinIO

S3-compatible object storage you host yourself. Same verb set as AWS S3, pointed at your own endpoint.

Field	Value
Plugin ID	minio
Category	storage
Auto Enable	No
Distribution	registry
Capabilities	storage, put-object, get-object, presigned-put

Settings (required: endpoint, bucket):

Setting	Type	Default	Environment variable	Description
endpoint	string	—	MINIO_ENDPOINT	Full endpoint URL, e.g. https://minio.example.com:9000
region	string	us-east-1	MINIO_REGION	Region label sent in S3 requests (MinIO ignores it)
bucket	string	—	MINIO_BUCKET	Bucket name
accessKey	string	—	MINIO_ACCESS_KEY	MinIO access key (secret)
secretKey	string	—	MINIO_SECRET_KEY	MinIO secret key (secret)
presignExpiresSeconds	number	600	MINIO_PRESIGN_EXPIRES_SECONDS	Pre-signed upload URL TTL, 60–3600

GitHub Storage

Stores uploads as files in a GitHub repository, with optional Git LFS. Keeps every asset inside the Git-native ownership model — the same repos you already own.

Field	Value
Plugin ID	github-storage
Category	storage
Auto Enable	No
Distribution	registry
Capabilities	storage, put-object, get-object, lfs

Modes (the mode setting):

Mode	Behaviour
separate-repo (default)	One operator-owned repository holds every upload. Owner and repo come from the settings UI (the same OAuth-connected selectors as work creation) or from the env vars.
data-repo	Uploads land in each Work's own data repo, resolved per upload from the Work's owner and storage config, authenticated with that Work owner's OAuth token.

data-repo mode requires the API to inject a WorkRepoResolver into the plugin context at boot (apps/api/src/uploads/storage-backend.factory.ts does this) and requires every upload to carry workId — anonymous uploads are rejected with a configuration error rather than silently misfiled.

Git LFS. With lfsEnabled on, the blob goes to GitHub's LFS storage via the LFS Batch API and only a pointer file is committed; the plugin also keeps an idempotent .gitattributes entry tracking <pathPrefix>/**. It defaults to true for fresh deployments and false for deployments that already had the legacy env vars set without an explicit mode, so existing setups keep a byte-for-byte identical commit shape. LFS deletes are best-effort — removing the pointer commit drops the file from the branch tree, but GitHub's public API exposes no LFS object purge, matching git lfs rm.

Transport setting	Values	Notes
lfsTransport	api (default), git-cli	api calls the LFS Batch API over HTTPS via the Octokit token — no binaries needed. git-cli shells out and needs git ≥ 2.40 + git-lfs ≥ 3.4 on PATH.
transport (non-LFS commit path)	auto (default), contents-api, clone-and-push	auto picks contents-api for separate-repo and clone-and-push (isomorphic-git) for data-repo.

Environment Variables:

Variable	Required	Description
GITHUB_STORAGE_MODE	No	separate-repo (default) or data-repo
GITHUB_STORAGE_TOKEN	Yes in separate-repo	PAT with contents:write on the storage repo
GITHUB_STORAGE_OWNER	Yes in separate-repo	Repository owner
GITHUB_STORAGE_REPO	Yes in separate-repo	Repository name
GITHUB_STORAGE_BRANCH	No	Default main
GITHUB_STORAGE_PATH_PREFIX	No	Default uploads
GITHUB_STORAGE_LFS_ENABLED	No	true / false — see the default rule above
GITHUB_STORAGE_LFS_TRANSPORT	No	api (default) or git-cli
GITHUB_STORAGE_TRANSPORT	No	auto / contents-api / clone-and-push
GITHUB_STORAGE_PUBLIC_URL_BASE	No	Public raw URL base (e.g. a CDN in front of a public repo); unset routes reads through the authenticated API

The matching settings keys are mode, token, owner, repo, branch, pathPrefix, lfsEnabled, lfsTransport and transport — the settings form and the environment variables above are two views of the same values.

Database

PostgreSQL DB

Holds the relational database connection used by deployed Works. Distinct from storage (object storage): this is the SQL server a Work's site talks to. The plugin owns the tenant-level backend choice and provisions or derives a database per Work.

Field	Value
Plugin ID	postgres-db
Category	database
Configuration Mode	hybrid
Auto Enable	Yes
System Plugin	Yes
Distribution	core
Capabilities	database, datastore

Settings:

Setting	Type	Scope	Default	Description
mode	string	user	ever-works-db	ever-works-db — a managed database provisioned for you, one per Work, no setup. custom — connect your own Postgres server.
customConnectionString	string	user	—	postgresql://user:password@host:5432/db, used as the server for all your Works. A database is created per Work when the role allows CREATE DATABASE; otherwise the connection is used as-is. Secret; stored encrypted and shown masked. Visible only when mode is custom.
overrideConnectionString	string	work	—	Override the database for THIS Work only. Secret; writable from the Work's Deploy page (/works/:id/deploy) — the scope guards enforce that. Blank falls back to the account-level setting.

See Managed Hosting for how the managed ever-works-db mode fits with *.ever.works subdomains.

Vector Store

Vector-store plugins hold the Knowledge Base chunk embeddings and answer retrieval queries. Every plugin normalises its raw vendor score into [0, 1] (higher is better) so the retrieval contract stays backend-independent, and declares how it isolates one Work from another (namespacePerWork).

pgvector

The default vector store. Keeps embeddings in the same Postgres the API already uses, in the work_knowledge_chunks table created by the agent migrations — zero extra infrastructure.

Field	Value
Plugin ID	pgvector
Category	vector-store
Default For	vector-store
Auto Enable	Yes
System Plugin	Yes
Distribution	core
Tenancy mode	rowFilter (per-Work via SQL)

Settings:

Setting	Type	Default	Environment variable	Description
embeddingModel	string	text-embedding-3-small	KB_EMBEDDING_MODEL	Must match the model that produced the existing rows — a mismatch retrieves from a mixed vector space and recall drops sharply
embeddingDimensions	number	1536	KB_EMBEDDING_DIMENSIONS	vector(N) column dimension (1–16000). Changing it needs a column-altering migration plus a full re-embed
indexType	string	ivfflat	KB_PGVECTOR_INDEX_TYPE	ivfflat or hnsw; hnsw is faster at query time but needs pgvector ≥ 0.5.0 and a separate index build
lists	number	100	KB_PGVECTOR_LISTS	ivfflat inverted lists — tune to about sqrt(rows)
efSearch	number	40	KB_PGVECTOR_EF_SEARCH	HNSW recall-vs-latency knob; ignored when indexType is ivfflat

Every retrieval applies WHERE work_id = $1, so per-Work isolation holds even though all Works share one table.

Qdrant

Stores embeddings in a Qdrant cluster (Qdrant Cloud or self-hosted) with one collection per Work — deleteByWork becomes a single collection drop instead of a whole-index payload-filter delete, and HNSW parameters can be tuned per tenant.

Field	Value
Plugin ID	qdrant
Category	vector-store
Auto Enable	No
Distribution	registry (install-on-demand)
Tenancy mode	collection (one per Work)
Supports filter	Yes (payload filter pushdown)
Supports hybrid	No (vector-only retrieval)

Settings:

Setting	Type	Default	Environment variable	Description
qdrantUrl	string	http://localhost:6333	QDRANT_URL	HTTP(S) endpoint of the Qdrant instance
qdrantApiKey	string	—	QDRANT_API_KEY	Secret. Required for Qdrant Cloud and any cluster behind auth
collectionPrefix	string	ever-works-kb	QDRANT_COLLECTION_PREFIX	Final collection name is {prefix}-{workId}
embeddingModel	string	text-embedding-3-small	KB_EMBEDDING_MODEL	Must match the model that produced the points already in the collection
vectorSize	number	1536	QDRANT_VECTOR_SIZE	Changing it requires re-creating the collection
distance	string	cosine	QDRANT_DISTANCE	cosine, dot, or euclid — drives both the collection config and the score normalisation
upsertBatchSize	number	128	QDRANT_UPSERT_BATCH_SIZE	Points per POST /points request

Score normalisation branches on the distance metric: cosine maps (raw + 1) / 2, dot applies a sigmoid, euclid uses 1 / (1 + raw) — every branch clamped to [0, 1].

Pipeline

The platform ships 14 pipeline plugins — three first-party engines plus eleven that delegate generation to an external agent or automation platform. See AI & Generation for a comparison.

Standard Pipeline

The default 15-step structured generation pipeline. Uses LangChain for AI operations with configurable search, extraction, and content generation steps.

Field	Value
Plugin ID	standard-pipeline
Configuration Mode	default
Auto Enable	Yes
Default For	pipeline
System Plugin	Yes
Capabilities	pipeline, form-schema-provider

The 15 pipeline steps, organized into 8 phases:

1. Prompt Comparison
2. Prompt Processing
3. Domain Detection
4. AI First Items Generation
5. Search Queries Generation
6. Web Search
7. Content Retrieval
8. Content Filtering
9. Items Extraction
10. Deduplication and Data Aggregation
11. Categories and Tags Processing
12. Sources Validation
13. Badges Processing
14. Image Capture
15. Markdown Generation

Agent Pipeline

Autonomous AI agent pipeline using the Vercel AI SDK with tool calling. The agent independently researches and generates work items using a parent/worker model architecture.

Field	Value
Plugin ID	agent-pipeline
Configuration Mode	hybrid
Auto Enable	No
Capabilities	pipeline, form-schema-provider

Settings:

Setting	Type	Default	Description
maxSteps	integer	50	Maximum agent tool-calling steps (10–2000)

Claude Code Generator

Generation pipeline that uses the Claude Code CLI to autonomously research and generate work items. Requires either an OAuth token or Anthropic API key.

Field	Value
Plugin ID	claude-code
Configuration Mode	user-required
Auto Enable	No
Capabilities	pipeline, form-schema-provider

Settings:

Setting	Type	Default	Description
oauthToken	string	—	Claude Code OAuth token (secret, from claude setup-token)
apiKey	string	—	Anthropic API key (secret, alternative to OAuth)
model	string	—	Model alias or full name (e.g., sonnet, opus)
version	string	2.1.37	Claude Code CLI version
maxTurns	number	500	Maximum conversation turns (1–100)
maxBudgetUsd	number	—	Maximum budget in USD

Environment Variables:

Variable	Required	Description
PLUGIN_CLAUDE_CODE_OAUTH_TOKEN	No	OAuth token (can be set in user settings instead)

Claude Managed Agent

Hosted Claude Managed Agent pipeline. Delegates the full work generation to Anthropic's managed agent runtime instead of orchestrating the steps locally.

Field	Value
Plugin ID	claude-managed-agent
Configuration Mode	user-required
Auto Enable	No
Capabilities	pipeline, form-schema-provider

See Claude Managed Agent Plugin for setup, settings, and the full list of environment variables. Refer to packages/plugins/claude-managed-agent/src/ for the latest schema.

Codex Generator

Pipeline plugin that delegates the full generation to OpenAI Codex. Useful when you want Codex's tool-using behaviour to drive the entire generation flow.

Field	Value
Plugin ID	codex
Configuration Mode	user-required
Auto Enable	No
Capabilities	pipeline, form-schema-provider

See Codex Plugin for setup. Refer to packages/plugins/codex/src/ for the current settings schema.

Gemini Generator

Pipeline plugin that delegates the full generation to the Gemini CLI agent. Distinct from the google AI provider plugin: Gemini Generator runs as an autonomous CLI-driven pipeline, while the google plugin exposes Gemini models for use as a regular AI provider in the Standard or Agent pipelines.

Field	Value
Plugin ID	gemini
Configuration Mode	user-required
Auto Enable	No
Capabilities	pipeline, form-schema-provider

See Gemini Plugin for setup. Refer to packages/plugins/gemini/src/ for the current settings schema.

OpenCode Generator

Pipeline plugin that delegates the full generation to OpenCode, an open-source code agent.

Field	Value
Plugin ID	opencode
Configuration Mode	user-required
Auto Enable	No
Capabilities	pipeline, form-schema-provider

See OpenCode Plugin for setup. Refer to packages/plugins/opencode/src/ for the current settings schema.

Make.com Workflows

Pipeline plugin that triggers Make.com (formerly Integromat) scenarios via webhooks to handle work generation. Use this to plug in a no-code/low-code workflow as the source of generated items.

Field	Value
Plugin ID	make
Configuration Mode	user-required
Auto Enable	No
Capabilities	pipeline, form-schema-provider

See Make.com Plugin for setup. Refer to packages/plugins/make/src/ for the current settings schema.

SIM AI Workflows

Pipeline plugin that delegates work generation to a SIM AI workflow defined in the SIM Studio platform.

Field	Value
Plugin ID	sim-ai
Configuration Mode	user-required
Auto Enable	No
Capabilities	pipeline, form-schema-provider

See SIM AI Workflows Plugin for setup. Refer to packages/plugins/sim-ai/src/ for the current settings schema.

Zapier Automation

Pipeline plugin that triggers Zapier actions during work generation. Lets you wire generation events to any of Zapier's 7000+ integrations.

Field	Value
Plugin ID	zapier
Configuration Mode	user-required
Auto Enable	No
Capabilities	pipeline, form-schema-provider

See Zapier Plugin for setup. Refer to packages/plugins/zapier/src/ for the current settings schema.

Composio Integrations

Pipeline plugin that executes Composio tools during work generation. Gives Ever Works access to 500+ third-party app integrations (Gmail, Slack, GitHub, Notion, Linear, Salesforce, …) with OAuth brokered per user by Composio.

Field	Value
Plugin ID	composio
Configuration Mode	user-required
Auto Enable	No
Capabilities	pipeline, form-schema-provider

See Composio Plugin for setup. Refer to packages/plugins/composio/src/ for the current settings schema.

Activepieces Automation

Pipeline plugin that delegates work generation to Activepieces flows. It triggers a flow webhook at the execute stage and collects structured items from the flow's Return Response action — an AI-first, open-source alternative to Make.com and Zapier that you can self-host or run on Activepieces Cloud.

Field	Value
Plugin ID	activepieces
Configuration Mode	user-required
Auto Enable	No
Capabilities	pipeline, form-schema-provider

See Activepieces Plugin for setup. Refer to packages/plugins/activepieces/src/ for the current settings schema.

Hermes Agent

Pipeline plugin that uses a preconfigured Hermes Agent installation on the API host as the generation engine. Ever Works provisions an isolated workspace for the run, launches Hermes in one-shot CLI mode against it, and validates the structured result file Hermes writes back before storing the generated items.

Field	Value
Plugin ID	hermes-agent
Category	pipeline
Configuration Mode	user-required
Auto Enable	No
Distribution	registry
Capabilities	pipeline, form-schema-provider

Settings (required: profile):

Setting	Type	Default	Description
profile	string	default	Hermes profile already configured on the backend via hermes model (user-scoped)
provider	string	—	Optional Hermes provider override passed to the CLI for each run
model	string	—	Optional Hermes model override passed to the CLI for each run
toolsets	string	web,skills	Comma-separated Hermes toolsets to enable for generation
skills	string	—	Optional comma-separated Hermes skills to preload
maxTurns	number	90	Maximum Hermes tool-calling turns per run (1–500)
yolo	boolean	true	Bypass Hermes approval prompts so unattended runs do not stall
binaryPath	string	hermes	CLI executable path, resolved against the host PATH (hidden; env PLUGIN_HERMES_BINARY_PATH)

Backend prerequisites:

1. Install Hermes Agent on the machine running the Ever Works API.
2. Run hermes model for the profile you intend to use.
3. Enter that profile name in the plugin's settings page, then select hermes-agent as the pipeline for the Work at /works/:id/plugins.

Hermes provider secrets stay in Hermes' own profile configuration — this plugin does not manage them. Hermes runs do not persist checkpoints (only standard-pipeline does), so a host restart mid-run means cancelling and re-triggering generation.

Go-to-Market Pipeline

The engine behind Campaign Works. Runs the go-to-market stage set — research → qualify → draft → review → act → follow-up → enrich → measure — with every stage declaring its requires / provides keys so hand-offs are explicit and auditable. Review is a human gate placed before any outbound action, and the act stage only stages approved drafts for delivery: it never sends.

Field	Value
Plugin ID	gtm-pipeline
Category	pipeline
Auto Enable	No
Distribution	registry
Capabilities	pipeline, form-schema-provider

Generator form fields (resolved from the Work's generator config; out-of-range values clamp to the default):

Field	Type	Default	Range / values	Description
target_channels	tags	email	email, blog, social, newsletter, community	Channels to prepare content for
tone	select	professional	professional, friendly, direct, enthusiastic	Voice used for drafted content
cadence	select	weekly	daily, weekly, biweekly, monthly	How often the pipeline is expected to run
max_contacts_per_run	number	20	1–200	Upper bound of contacts drafted per run
qualify_min_score	number	40	0–100	Minimum qualify score required to keep a contact
risk_exclude_threshold	number	7	0–10	Contacts at or above this risk score are excluded
review_required	bool	true	—	Human gate before any outbound action (drafts-not-sends)
follow_up_quiet_days	number	4	1–90	Days of silence after which a prepared action queues a follow-up

Creating a Campaign Work at /works/new/campaign provisions the Work, its Goal, the GTM Agents and the seeded Tasks in one call, with this pipeline already selected. See Campaigns.

Prompt Management

Langfuse

External prompt management plugin. Lets you store, version, label, and A/B-test all pipeline prompts in Langfuse instead of shipping them in-repo.

Field	Value
Plugin ID	langfuse
Category	utility
Auto Enable	Yes
Default For	prompt-provider
Distribution	registry
Capabilities	prompt-provider

Environment Variables:

Variable	Required	Description
PLUGIN_LANGFUSE_SECRET_KEY	No	Langfuse secret key for API authentication
PLUGIN_LANGFUSE_PUBLIC_KEY	No	Langfuse public key for API authentication
PLUGIN_LANGFUSE_BASE_URL	No	Langfuse base URL (for self-hosted instances)

See Langfuse Plugin for setup, label conventions, and fallback behaviour. Refer to packages/plugins/langfuse/src/ for the current settings schema.

Email Providers

Email-provider plugins implement email-outbound and, where the vendor supports it, email-inbound. They send every platform email — notifications, digests, and Agent mailboxes — and parse replies back in. Configure them at Settings → Integrations → Emails (/settings/integrations/emails). Every provider de-dupes on EmailSendInput.messageRef, so a retried send is never delivered twice.

Plugin ID	Capabilities	Transport
resend	email-outbound	Official resend SDK with native idempotency keys
sendgrid	email-outbound	Official @sendgrid/mail SDK, v3 Mail Send
mailchimp-transactional	email-outbound	Official @mailchimp/mailchimp_transactional SDK (Mandrill)
mailgun	email-outbound, email-inbound	Official mailgun.js SDK + signed inbound routes
postmark	email-outbound, email-inbound	Postmark Server API + Inbound Streams webhook parser

Settings by provider:

Plugin	Setting	Required	Secret	Environment variable	Notes
resend	apiKey	Yes	Yes	RESEND_API_KEY	Resend API key
resend	defaultSenderDomain	No	No	—	Default from domain
sendgrid	apiKey	Yes	Yes	SENDGRID_API_KEY	A fresh MailService per send keeps the key request-scoped
sendgrid	defaultSenderDomain	No	No	—	Default From domain
mailchimp-transactional	apiKey	Yes	Yes	MANDRILL_API_KEY	Transactional (Mandrill) product, not the Marketing API
mailchimp-transactional	defaultSenderDomain	No	No	—	Default From domain
mailgun	apiKey	Yes	Yes	MAILGUN_API_KEY	—
mailgun	domain	Yes	No	MAILGUN_DOMAIN	Sending domain
mailgun	region	No	No	MAILGUN_REGION	us (default) or eu
mailgun	webhookSigningKey	No	Yes	MAILGUN_WEBHOOK_SIGNING_KEY	Enables inbound HMAC-SHA256 verification; unset skips it
postmark	apiKey	Yes	Yes	POSTMARK_API_KEY	Postmark Server API token
postmark	defaultSenderDomain	No	No	—	Default From domain
postmark	inboundWebhookSecret	No	Yes	POSTMARK_INBOUND_SECRET	Basic-Auth secret for inbound webhook verification
postmark	inboundStreamId	No	No	—	Specific inbound stream id

Inbound webhook URLs (register these in the provider's dashboard):

Provider	Purpose	URL
Postmark	Inbound email	https://<your-domain>/api/email/inbound/postmark
Postmark	Delivery events	https://<your-domain>/api/email/events/postmark

Postmark surfaces Delivery, Bounce, SpamComplaint, Open and Click events. Mailgun verifies inbound with HMAC(signingKey, timestamp + token) compared in constant time and decodes both JSON and form-urlencoded webhook bodies. Resend inbound is not supported yet — Resend's inbound product is still in private beta.

Notification Channels

Notification-channel plugins are outbound-only delivery surfaces for platform notifications. Add and test them at Settings → Integrations → Channels (/settings/integrations/channels); each channel row has a Send test action backed by POST /api/notification-channels/:id/test. See Notifications.

Each plugin declares a shape — broadcast (a room everyone sees), direct (one recipient), or workflow (the vendor fans out itself) — plus per-channel targetConfig and tenant-wide default settings.

Plugin ID	Shape	Transport
slack-channel	broadcast	Official @slack/webhook SDK over an incoming webhook
discord-channel	broadcast	Discord incoming webhook URL (plain HTTPS POST)
telegram-channel	direct	Official grammy SDK (Api.sendMessage)
whatsapp-channel	direct	WhatsApp Business Cloud API (/{phoneNumberId}/messages)
novu-channel	workflow	Novu Trigger API (POST /v1/events/trigger)

Per-channel targetConfig:

Plugin	Key	Required	Description
slack-channel	webhookUrl	Yes	Slack incoming webhook URL (https://hooks.slack.com/...)
slack-channel	username	No	Override sender username for this channel
slack-channel	iconEmoji	No	Override sender icon emoji for this channel
discord-channel	webhookUrl	Yes	Discord channel incoming webhook URL
discord-channel	username	No	Override sender username for this channel
discord-channel	avatarUrl	No	Override sender avatar for this channel
telegram-channel	botToken	Yes	Bot token from @BotFather
telegram-channel	chatId	Yes	Destination chat id
whatsapp-channel	accessToken	Yes	Meta system-user access token
whatsapp-channel	phoneNumberId	Yes	WhatsApp Business phone-number id
whatsapp-channel	to	Yes	Recipient phone in E.164, e.g. +15551234567
novu-channel	apiKey	Yes	Novu API key (environment-scoped)
novu-channel	workflowId	Yes	Novu workflow trigger identifier (the workflow's name field)
novu-channel	subscriberId	Yes	Novu subscriber to deliver to

Plugin-level defaults (tenant-wide fallbacks used when a channel does not override them):

Plugin	Setting	Description
slack-channel	defaultUsername	Fallback sender username
slack-channel	defaultIconEmoji	Fallback icon emoji, e.g. :robot_face:
discord-channel	defaultUsername	Fallback sender username
discord-channel	defaultAvatarUrl	Fallback sender avatar
telegram-channel	disableNotification	Send silently — no sound or vibration
whatsapp-channel	apiVersion	Graph API version, defaults to v21.0
novu-channel	apiBase	Defaults to https://api.novu.co; set for self-hosted or EU (https://eu.api.novu.co)

Rich payloads. Slack accepts Block Kit via the slack-blocks payload kind, Discord accepts embeds via discord-embeds, Telegram accepts MarkdownV2 via telegram-markdown, and WhatsApp accepts a pre-approved template via whatsapp-template ({ name, language, components }).

Finding a Telegram chat id:

1. Create a bot with @BotFather and copy the botToken.
2. Send any message to the bot, or add it to a group.
3. GET https://api.telegram.org/bot<botToken>/getUpdates and read result[].message.chat.id.
4. Paste that value into chatId — verifyTarget calls getMe before the first send to confirm the token.

WhatsApp's 24-hour window

WhatsApp only allows free-form text within 24 hours of the recipient's last message. Outside that window you must send a pre-approved template via the whatsapp-template payload kind; plain text is best-effort and in-window only.

Connectors

Connectors are bidirectional communication plugins: they send outbound messages or records and pull inbound activity into the event-ingest spine as IngestedEventEnvelopes. A connector is a superset of a notification-channel — both families coexist, and both are distinct from the third-party automation aggregators in the Pipeline section. Manage them at Settings → Plugins → Connector (/settings/plugins/connector); see Connectors and Integrations.

Every connector with the event-source capability supports an opt-in historical backfillDays window (default 0 = off, max 90) that widens the first pull only, bounded to a per-phase page cap. Re-delivery is free — the ingest pipeline dedupes on (source, sourceEventId). Connectors that also expose the optional backfill() method can import history at any time over an explicit window rather than only as a side effect of the first pull.

Plugin ID	Capabilities	SDK	Outbound
slack-connector	connector, connector-slack, event-source	@slack/web-api	chat.postMessage with a bot token
discord-connector	connector, connector-discord	discord.js REST client	POST /channels/:id/messages with a bot token
linear-connector	connector, connector-linear, event-source	@linear/sdk	Comment on a Linear issue
notion-connector	connector, connector-notion, event-source	@notionhq/client	Comment on a Notion page
jira-connector	connector, event-source	jira.js	Comment on a Jira issue
hubspot-connector	connector, connector-hubspot, event-source	@hubspot/api-client	Note engagement on a CRM record
pipedrive-connector	connector, connector-pipedrive, event-source	pipedrive	Note on a deal, person, or organization
google-workspace-connector	connector, event-source	@googleapis/drive, @googleapis/calendar	Drive + Calendar sweep (ingest-focused)
zoom-connector	connector, event-source	@zoom/rivet	Cloud-recording + transcript sweep
bluesky-connector	connector, connector-bluesky, event-source	@atproto/api	Post, or a threaded reply
mastodon-connector	connector, connector-mastodon, event-source	masto	Status at the configured visibility

Settings by connector:

Plugin	Setting	Secret	Environment variable	Description
slack-connector	botToken	Yes	SLACK_BOT_TOKEN	Bot User OAuth token (xoxb-…)
slack-connector	signingSecret	Yes	SLACK_SIGNING_SECRET	For the inbound Events API
slack-connector	appId	No	—	Slack app id
slack-connector	defaultChannelId	No	—	Default destination channel, e.g. C0123456789
slack-connector	eventChannelIds	No	—	Channels to ingest events from (comma-separated ids); defaults to the default channel
discord-connector	botToken	Yes	DISCORD_BOT_TOKEN	Bot token
discord-connector	publicKey	Yes	DISCORD_PUBLIC_KEY	Application public key for the inbound Interactions API
discord-connector	applicationId	No	—	Discord application (client) id
discord-connector	guildId	No	—	Default guild/server id
discord-connector	defaultChannelId	No	—	Default destination channel
linear-connector	apiKey	Yes	LINEAR_API_KEY	Linear API key (lin_api_…)
linear-connector	teamIds	No	—	Comma-separated team-id filter for the pull
linear-connector	defaultIssueId	No	—	Default issue for outbound comments
notion-connector	apiKey	Yes	NOTION_API_KEY	Notion integration token
notion-connector	databaseIds	No	—	Comma-separated database filter; empty falls back to workspace search
notion-connector	defaultPageId	No	—	Default page for outbound comments
jira-connector	baseUrl	No	JIRA_BASE_URL	Jira site URL, https:// only — validated against SSRF before every call
jira-connector	email	No	JIRA_EMAIL	Atlassian account email (API-token basic auth)
jira-connector	apiToken	Yes	JIRA_API_TOKEN	Atlassian API token
jira-connector	projectKeys	No	—	Comma-separated project-key filter, whitelisted against Jira's key alphabet
jira-connector	defaultIssueKey	No	—	Default issue for outbound comments
hubspot-connector	accessToken	Yes	HUBSPOT_ACCESS_TOKEN	Private-app token
hubspot-connector	objectTypes	No	—	Comma-separated sweep list; contacts, companies, deals when empty
hubspot-connector	portalId	No	—	Portal (hub) id — enables record deep links on every envelope
hubspot-connector	defaultObjectType	No	—	Default object type for createRecord and the verify probe
hubspot-connector	defaultAssociatedObjectId	No	—	Default CRM record outbound notes attach to
pipedrive-connector	apiToken	Yes	PIPEDRIVE_API_TOKEN	API token
pipedrive-connector	entityTypes	No	—	Comma-separated sweep list; deals, persons, organizations by default
pipedrive-connector	companyDomain	No	—	acme for acme.pipedrive.com — enables record deep links
pipedrive-connector	defaultDealId	No	—	Default deal outbound notes attach to
google-workspace-connector	clientId	No	GOOGLE_WORKSPACE_CLIENT_ID	Google OAuth client id
google-workspace-connector	clientSecret	Yes	GOOGLE_WORKSPACE_CLIENT_SECRET	OAuth client secret
google-workspace-connector	refreshToken	Yes	GOOGLE_WORKSPACE_REFRESH_TOKEN	OAuth refresh token
google-workspace-connector	surfaces	No	—	drive, calendar, or both (default both)
google-workspace-connector	driveFolderIds	No	—	Comma-separated Drive folder filter
google-workspace-connector	calendarIds	No	—	Comma-separated calendar ids, default primary
google-workspace-connector	meetTranscripts	No	—	Export Meet transcript docs into meeting envelopes (default true)
zoom-connector	accountId	No	ZOOM_ACCOUNT_ID	Zoom account id of the Server-to-Server OAuth app (required)
zoom-connector	clientId	No	ZOOM_CLIENT_ID	Client id of the Server-to-Server OAuth app (required)
zoom-connector	clientSecret	Yes	ZOOM_CLIENT_SECRET	Client secret of the Server-to-Server OAuth app (required)
bluesky-connector	identifier	No	—	Handle or DID of the connected account
bluesky-connector	appPassword	Yes	BLUESKY_APP_PASSWORD	Always an app password, never the account password
bluesky-connector	service	No	—	PDS URL, defaults to https://bsky.social, SSRF-guarded
mastodon-connector	instanceUrl	No	—	Instance base URL, SSRF-guarded before every call
mastodon-connector	accessToken	Yes	MASTODON_ACCESS_TOKEN	Application token
mastodon-connector	defaultVisibility	No	—	public, unlisted, private, or direct

Every event-source connector also accepts backfillDays (0–90) as described above. Required settings, from each plugin's settingsSchema.required: botToken (Slack, Discord); apiKey (Linear, Notion); apiToken (Pipedrive); accessToken (HubSpot); baseUrl + email + apiToken (Jira); clientId + clientSecret + refreshToken (Google Workspace); accountId + clientId + clientSecret (Zoom); identifier + appPassword (Bluesky); instanceUrl + accessToken (Mastodon). Per-send overrides such as Slack's and Discord's channelId, Linear's issueId, Notion's pageId or HubSpot's associatedObjectId travel on the send's targetConfig, not on plugin settings.

Inbound status. The Slack and Discord connectors ship the outbound leg first; inbound message routing (signature verify → pair → route to an Agent → reply in-thread) is a documented follow-up in each plugin's README, with the signingSecret / publicKey settings already captured so no reconfiguration is needed when it lands.

Metrics

Metrics plugins implement the read-only metrics-provider capability. The MetricsFacadeService in @ever-works/agent routes listMetrics and getMetricValue through them so Goals can evaluate targets like "keep signups above 100" without any vendor being hard-coded into the platform.

Plugin ID	Reads	Notes
stripe-metrics	Stripe balance and income via the official stripe SDK	Values reported in the configured currency
google-analytics-metrics	GA4 Data API via @google-analytics/data (runReport)	Active users, conversions, and other GA4 metrics
posthog-metrics	PostHog Query API	posthog-node is ingestion-only, so this plugin uses the documented Query API
custom-http-metrics	Any JSON HTTP endpoint you control	One metric per configured endpoint, supportedWindows: ['point']

Settings:

Plugin	Setting	Required	Default	Environment variable	Description
stripe-metrics	secretKey	Yes	—	STRIPE_SECRET_KEY	Stripe secret key (a restricted rk_... key is recommended)
stripe-metrics	currency	No	usd	—	Lowercase ISO-4217 code metric values are reported in
google-analytics-metrics	propertyId	Yes	—	GOOGLE_ANALYTICS_PROPERTY_ID	Numeric GA4 property id; properties/123456789 also accepted
google-analytics-metrics	serviceAccountJson	Yes	—	GOOGLE_ANALYTICS_SERVICE_ACCOUNT_JSON	Full JSON service-account key file (secret)
posthog-metrics	projectId	Yes	—	POSTHOG_PROJECT_ID	Numeric project id
posthog-metrics	personalApiKey	Yes	—	POSTHOG_PERSONAL_API_KEY	Personal API key (secret)
posthog-metrics	apiHost	No	https://us.posthog.com	—	EU Cloud is https://eu.posthog.com; self-hosted URLs work too
custom-http-metrics	endpoints	Yes	—	—	Array of endpoint definitions (below)

A custom-http-metrics endpoint entry:

{
	"endpoints": [
		{
			"id": "mrr", // stable metric id, referenced by Goals
			"label": "Monthly recurring revenue",
			"url": "https://metrics.example.com/mrr",
			"unit": "usd", // optional, defaults to "count"
			"valuePath": "data.metrics[0].value",
			"method": "GET", // optional; GET is the only allowed value
			"headers": {
				// optional; values are stored as secrets
				"Authorization": "Bearer …"
			}
		}
	]
}

DNS

Cloudflare DNS

Creates, updates, and removes DNS records for the subdomains and custom domains your Works are served from. Implements IDnsProvider and runs alongside the legacy concrete provider in packages/agent/src/ever-works-providers/cloudflare-dns.provider.ts — this plugin is additive, not a replacement.

Field	Value
Plugin ID	cloudflare-dns
Category	dns
Auto Enable	No
Visibility	user-only
Distribution	registry
Capabilities	dns, dns-ensure-record, dns-remove-record, dns-record-exists, dns-root-domain

Two modes, side by side:

Mode	Who configures it	Where credentials live
Managed (*.ever.works)	Platform operator	Environment variables — the x-envVar schema entries forward them into plugin settings, so no per-user setup is needed.
Bring your own (your apex)	The account owner	Encrypted user-scoped plugin settings. Records are created with the Cloudflare proxy off so you keep serving your TLS.

Settings (required: apiToken, zoneId):

Setting	Type	Default	Environment variable	Description
apiToken	string	—	CLOUDFLARE_API_TOKEN	Scoped token with DNS:Edit on the target zone (secret, user-scoped)
zoneId	string	—	CLOUDFLARE_ZONE_ID	Zone id that owns the root domain
rootDomain	string	ever.works	EVER_WORKS_DOMAIN	The DNS zone this provider manages
targetHostname	string	—	EVER_WORKS_DEPLOY_LB_HOSTNAME	CNAME target — the ingress load-balancer hostname public Work subdomains resolve to (admin-only)
proxied	boolean	true	—	Create records behind the Cloudflare proxy (Universal SSL out of the box); set false for custom domains you already terminate TLS for

Create the token at https://dash.cloudflare.com/profile/api-tokens with DNS:Edit on the target zone.

Capability methods:

Capability	Method	Behaviour
dns-ensure-record	ensureRecord({ host, type, target, proxied? })	Idempotent create-or-update; patches drifted records in place
dns-remove-record	removeRecord({ host, type? })	Idempotent delete; omitting type probes both CNAME and A
dns-record-exists	recordExists(host)	Uniqueness probe — true if any CNAME or A record exists for the host
dns-root-domain	rootDomain()	Returns the zone's root domain

See Custom Domains and Managed Hosting.

Identity

OpenID Connect identity (Ever ID)

Adds Sign in with Ever ID as an additional sign-in method, next to every existing one: e-mail and password, magic link, GitHub and Google keep working exactly as before. It is an OpenID Connect relying party (authorization code flow with PKCE S256), works with any standards-compliant provider, and never stores a provider token. See Ever ID for what people see.

Field	Value
Plugin ID	oidc-identity
Category	identity
Auto Enable	No — off until a platform administrator turns it on
Configured	Platform-wide only (admin-only)
Distribution	registry
Capabilities	identity-provider

Off by default, and quiet while off. Configuring the plugin does not turn anything on. Until a platform administrator runs Test connection and turns Ever ID on (on the administration page, /settings/admin/ever-id), the sign-in button is not shown, every sign-in route answers 404 and the API makes no request to the provider. Turning it off again takes effect within seconds on every API replica; people can still list and disconnect connected identities, and sign-out notices from the provider are still honoured.

Settings (required: issuerUrl, clientId, clientSecret):

Setting	Type	Default	Environment variable	Description
issuerUrl	string	—	EVER_ID_ISSUER_URL	The provider's issuer (https; http://localhost only outside production)
clientId	string	—	EVER_ID_CLIENT_ID	The client this installation is registered as
clientSecret	string	—	EVER_ID_CLIENT_SECRET	Client secret (client_secret_basic); write-only, never returned
allowedIssuers	array	[issuer]	EVER_ID_ALLOWED_ISSUERS	1–3 exact issuer strings accepted at once
apiAudience	string	ever-works	EVER_ID_API_AUDIENCE	The audience delegated and terminal access tokens must carry
signUpAllowed	boolean	true	EVER_ID_SIGN_UP_ALLOWED	Whether an unknown, verified identity may create an account after confirming
clockSkewSeconds	integer	60	EVER_ID_CLOCK_SKEW_SECONDS	Tolerance for token time checks (0–120)
localClients	array	[]	—	Up to 5 public clients (cli / node) allowed to exchange a device sign-in for a session
delegatedClientNames	array	[]	—	Names shown for apps that read a person's App Works with a delegated permission
accountManagementUrl	string	—	—	Target of the Manage in Ever ID link on the Connected identities card
displayName	string	Ever ID	—	The button and heading label

The last four are managed on the administration page (or PATCH /api/auth/ever-id/admin/settings); the environment-bound settings are operator configuration.

Register at the provider: one redirect address, <WEB_URL>/api/auth/ever-id/callback; the sign-out return address <WEB_URL>/api/auth/ever-id/logout-return; and, for sign-out notices, the back-channel logout address <API_URL>/api/auth/ever-id/backchannel-logout.

Job Runtimes

Job-runtime plugins implement IJobRuntimeProvider — the seam every background job in the platform goes through. One provider is active per instance, selected with EVER_WORKS_JOB_RUNTIME; the rest stay loaded but inert so a hot swap stays cheap. Tenants can override the instance default from Settings → Job Runtime (/settings/job-runtime), bounded by the operator allow-list EVER_WORKS_TENANT_RUNTIME_ALLOWED_PROVIDERS. See Job Runtimes.

All six declare the same capability set — job-runtime-enqueue, job-runtime-cancel, job-runtime-status, job-runtime-schedule, job-runtime-bind-tenant — and all six ship as core (bundled in every image). job-runtime-node additionally declares job-runtime-worker-host.

Plugin ID	EVER_WORKS_JOB_RUNTIME	Broker	Peer dependency the operator installs	Per-tenant isolation
job-runtime-trigger	trigger	Trigger.dev (push)	@trigger.dev/sdk	Per-tenant projectAccessToken
job-runtime-bullmq	bullmq	Redis	bullmq, ioredis	Redis queuePrefix per tenant
job-runtime-pgboss	pgboss	Postgres	pg-boss	Postgres schema per tenant
job-runtime-temporal	temporal	Temporal	@temporalio/client, @temporalio/worker	Temporal namespace per tenant
job-runtime-inngest	inngest	Inngest (serverless)	inngest	Per-tenant eventKey + signingKey
job-runtime-node	node	Your enrolled Fleet	none	Structural — the lease query only returns the owner's own work

Environment variables by runtime:

Runtime	Variables
trigger	TRIGGER_SECRET_KEY (prod tr_prod_*), TRIGGER_PROJECT_REF, optional TRIGGER_API_URL for self-hosted
bullmq	BULLMQ_REDIS_URL, BULLMQ_QUEUE_PREFIX
pgboss	PGBOSS_CONNECTION_STRING, PGBOSS_SCHEMA
temporal	TEMPORAL_ADDRESS, TEMPORAL_NAMESPACE, TEMPORAL_TLS_CERT + TEMPORAL_TLS_KEY (mTLS recommended)
inngest	INNGEST_EVENT_KEY, INNGEST_SIGNING_KEY
node	FLEET_NODE_API_URL, FLEET_NODE_LEASE_TTL_SECONDS, FLEET_NODE_REQUIRED_CAPABILITIES, FLEET_NODE_AGENT_TASK_COMMAND, FLEET_NODE_AGENT_TASK_WORKSPACE, FLEET_NODE_AGENT_TASK_ENV_PASSTHROUGH

None of the plugin packages depend on their broker SDK: operators pin and inject the real dispatchers themselves, and the shipped defaults are throwing stubs so a half-configured deployment fails loudly at first dispatch instead of silently dropping work. job-runtime-trigger and job-runtime-inngest keep startWorkerHost a deliberate no-op — both are push/serverless models where the vendor invokes your deployed code, so there is no worker process to start.

The node runtime — the queue is the fleet. Instead of an external broker, enqueue writes a lease-able fleet_jobs row and the machines you enrolled in Fleet poll for it over the same outbound-only HTTP channel enrollment and heartbeat already use — no inbound port is ever opened on a user's machine.

A lease is a deadline, not a lock: if a node dies mid-job nothing has to notice, because leaseExpiresAt passes and the work returns to the pool (or fails once the attempt budget is spent). Claiming is a conditional UPDATE pinned to status = 'queued', so two nodes racing the same row produce exactly one winner. JobEnqueueOptions.tags entries prefixed cap: become real scheduling requirements — a node may only lease a job whose every required tag is in its advertised capability set — while ordinary tags stay observability labels and never narrow eligibility. registerSchedules is intentionally a no-op: recurrence belongs to the platform cron, because anchoring a wall-clock schedule to intermittently-online machines would elect "whichever node happened to be awake" as the only one that ever fires.

Secret Stores

Secret-store plugins implement the secret-store-resolve capability. They turn an opaque credentialsSecretRef pointer stored on a tenant row into a plaintext credential bag at the moment a job runtime needs it, so no third-party credential is ever persisted in the platform database. All seven ship as core. The defaults inline: and env: need no plugin at all — they are handled in-process by packages/agent/src/tasks/in-process-secret-store-resolver.service.ts. See Secret Stores.

Each plugin claims exactly one pointer scheme and fails open — an unrecognised scheme or any error returns null with a warning rather than throwing.

Plugin ID	Backend	Pointer format	Example
secret-store-vault	HashiCorp Vault KV	vault:<path after /v1/>	vault:secret/data/tenants/acme/trigger
secret-store-k8s	Kubernetes Secrets	k8s:<name> or k8s:<namespace>/<name>	k8s:tenant-acme-creds
secret-store-aws-sm	AWS Secrets Manager	aws-sm:<region>/<secretName>	aws-sm:us-east-1/prod/tenants/acme
secret-store-gcp-sm	GCP Secret Manager	gcp-sm:<projectId>/<secretName>	gcp-sm:my-project/tenant-acme
secret-store-azure-kv	Azure Key Vault	azure-kv:<vault>/<secretName>	azure-kv:my-vault/prod-tenant-acme
secret-store-doppler	Doppler	doppler:<project>/<config>	doppler:ever-works/prd_tenants_acme
secret-store-infisical	Infisical	infisical:<workspaceId>/<environment>/<secretPath>	infisical:ws-abc/prod/tenants/acme

Settings (all resolved from environment variables):

Plugin	Setting	Secret	Environment variable	Description
secret-store-vault	vaultAddr	No	VAULT_ADDR	Vault server URL, e.g. https://vault.internal:8200
secret-store-vault	vaultToken	Yes	VAULT_TOKEN	Token with read permission on the requested paths
secret-store-k8s	kubernetesServiceHost	No	KUBERNETES_SERVICE_HOST	In-cluster API server IP, set by the kubelet automatically
secret-store-k8s	kubernetesServicePort	No	KUBERNETES_SERVICE_PORT	In-cluster API server port, defaults to 443
secret-store-aws-sm	awsAccessKeyId	No	AWS_ACCESS_KEY_ID	Omit to use the ambient AWS credential chain
secret-store-aws-sm	awsSecretAccessKey	Yes	AWS_SECRET_ACCESS_KEY	Omit to use the ambient AWS credential chain
secret-store-aws-sm	awsSessionToken	Yes	AWS_SESSION_TOKEN	For temporary STS credentials
secret-store-gcp-sm	gcpAccessToken	Yes	GCP_ACCESS_TOKEN	OAuth2 access token with the secretmanager.secretAccessor role
secret-store-azure-kv	azureKvToken	Yes	AZURE_KV_TOKEN	Azure AD bearer token, scope https://vault.azure.net/.default
secret-store-doppler	dopplerToken	Yes	DOPPLER_TOKEN	Service Token or Service Account token with read access
secret-store-infisical	infisicalToken	Yes	INFISICAL_TOKEN	Service Token or Machine Identity token with read access
secret-store-infisical	infisicalHost	No	INFISICAL_HOST	Base URL for self-hosted instances; defaults to https://app.infisical.com

The Vault resolver auto-detects the KV version: it tries KV v2 first (json.data.data is an object) and falls back to KV v1. The Kubernetes resolver reads its bearer token, CA cert and default namespace from the service-account mount at /var/run/secrets/kubernetes.io/serviceaccount/ and base64-decodes every Secret value; running out of cluster it returns null and warns, so local development should use inline: instead. GCP deliberately takes a pre-fetched GCP_ACCESS_TOKEN rather than signing service-account JWTs itself — operators provision it out of band via Workload Identity, a refresher sidecar, or a cron.

Utility

Comparison Generator

Auto-generates SEO-optimized A vs B comparison pages between work items.

Field	Value
Plugin ID	comparison-generator
Configuration Mode	hybrid
Auto Enable	No
System Plugin	Yes
Capabilities	form-schema-provider

Settings:

Setting	Type	Default	Description
cadence_override	string	use_work	Generation cadence: use_work, daily, weekly, monthly
max_comparisons_mode	string	custom	custom or unlimited
max_comparisons	number	50	Max total comparisons (1–500, only used in Custom mode)
min_items_for_comparison	number	3	Min items in category before generating (2–20)
ai_provider	string	—	Override AI provider for comparison generation
ai_model	string	—	Override AI model for comparison generation
custom_prompt	string	—	Additional instructions appended to comparison prompts
extended_analysis	boolean	false	Enable deep-dive 7-section extended analysis

See Comparisons for the full feature documentation.

Ever Works Skills

First-party skills-provider capability plugin (ADR-012). Sources the curated Skills catalog from the ever-works/skills GitHub repo (per ADR-014). Ships v1 with a built-in fallback catalog (cron-defaults / secret-handling / commit-message-style) so the plugin works before the upstream repo is created — the platform self-recovers when it appears.

Field	Value
Plugin ID	everworks-skills
Package	@ever-works/everworks-skills-plugin
License	MIT (ADR-014 catalog split)
Configuration Mode	admin-only
Auto Enable	Yes (default skills-provider)
Capabilities	skills-provider

Settings:

Setting	Type	Default	Description
catalogRepo	string	ever-works/skills	GitHub owner/repo of the catalog source.
catalogBranch	string	main	Branch to read from.
cacheTtlSeconds	number	3600	How long to cache the cloned catalog.

See Skills feature for the platform-side data model + resolver + injection pipeline.

Ever Works Playbooks

First-party playbook-provider capability plugin. Supplies the built-in Playbook catalogue shown in the Playbooks section of the capability catalogue (/catalog): eight packaged outcomes, each naming its trigger, its steps, the capabilities it needs (never a provider id), what it produces and when it stops to ask. Every entry is sanitised and validated before it is served, and five of the eight need no connection at all. PlaybookCatalogFacadeService merges the entries of every enabled playbook-provider; a later provider replaces an entry only with a strictly higher version.

Field	Value
Plugin ID	everworks-playbooks
Package	@ever-works/everworks-playbooks-plugin
License	MIT
Configuration Mode	admin-only
Auto Enable	Yes (default playbook-provider)
Distribution	registry
Capabilities	playbook-provider

No settings.

Ever Works Task Tracker

First-party task-tracker capability plugin (ADR-013). Thin shim over the platform's own DB-backed Tasks family — when this plugin is enabled, TasksFacadeService routes every Task operation through it. Community plugins (Linear / Jira / GitHub Issues) drop in by implementing the same ITaskTrackerPlugin contract.

Field	Value
Plugin ID	everworks-task-tracker
Package	@ever-works/everworks-task-tracker-plugin
License	MIT (ADR-014 catalog split)
Configuration Mode	admin-only
Auto Enable	No — declared default for task-tracker
Distribution	registry
Capabilities	task-tracker

Settings: none — the plugin binds to the platform's DB-backed service at boot via a runtime setPlatformTaskBackend() delegate.

See Task tracking feature for the full data model + state machine + chat thread + recurrence pipeline.

Agent Memory

First-party implementation of the agent-memory capability. Talks to a standalone agentmemory REST server — the same one Claude Code, Codex, OpenCode and MCP clients already use — so one memory store can be shared across every agent an operator runs. The AgentMemoryFacadeService dispatches to whichever agent-memory plugin the user or Work has resolved.

Field	Value
Plugin ID	agentmemory
Category	utility
Auto Enable	No
Distribution	registry
Capabilities	agent-memory

Settings:

Setting	Type	Scope	Default	Environment variable	Description
baseUrl	string	user	http://localhost:3111	AGENTMEMORY_BASE_URL	REST endpoint of the agentmemory server
apiKey	string	user	—	AGENTMEMORY_API_KEY	Bearer token (secret). Must match the server's AGENTMEMORY_SECRET; empty is fine for a localhost dev server
projectId	string	work	ever-works	AGENTMEMORY_PROJECT	Namespace sent as project on every request
timeoutMs	number	user	30000	—	Per-request timeout

Three ways to run the backend, same plugin code:

Mode	When	Setup
Local dev	Hacking on Ever Works on a laptop	npx @agentmemory/agentmemory in a second terminal — the default baseUrl already points at it
Self-hosted in cluster	Platform and memory store in the same Kubernetes namespace	Apply .deploy/k8s/agentmemory.optional.yaml and set baseUrl to http://agentmemory.<ns>.svc.cluster.local:3111
Hosted	You already run agentmemory elsewhere	Set baseUrl to the HTTPS endpoint and apiKey to the server's AGENTMEMORY_SECRET

The plugin uses only the documented REST subset — /agentmemory/health (which also drives validateConnection), /session/start, /session/end, /remember, /smart-search, /context, and /forget for governance deletes. See Memory.

Memory Pipeline Modifier

A pipeline-modifier plugin that injects two steps into whatever pipeline a Work runs: one at the start that fetches prior agent-memory context, and one at the end that saves a digest of what was generated. That is what makes consecutive generation runs build on each other instead of starting cold. Requires an agent-memory provider to be enabled.

Field	Value
Plugin ID	memory-pipeline-modifier
Category	utility
Auto Enable	No
Distribution	registry
Capabilities	pipeline-modifier

Settings (all work-scoped):

Setting	Type	Default	Description
enabled	boolean	false	Inject the memory hooks into this Work's pipeline
purpose	string	work-generation	Hint passed to the memory backend to bias retrieval, e.g. fix-bug or research
maxContextTokens	number	—	Upper bound on the injected memory context payload (100–32000)
saveSummary	boolean	true	Let the last pipeline step save a short observation about what was generated so the next run can recall it

The injected step ids are memory-fetch-context and memory-save.

Local Workspace

The workspace provider for self-hosted and desktop installs. Maintains a pool of real git worktree checkouts on the host so an Agent task gets an isolated working tree without a fresh clone every time. Each worktree carries a binding stamp written inside the gitdir, so it can never be committed and never appears in the working tree. See Task Isolation.

Field	Value
Plugin ID	local-workspace
Category	utility
Auto Enable	No
System Plugin	Yes
Distribution	core
Capabilities	workspace

Settings (all hidden/advanced):

Setting	Type	Default	Description
baseDir	string	EW_WORKSPACES_DIR or the OS temp dir	Directory the worktree pool lives under
fetchDepth	number	1	Shallow fetch depth for the base-ref fetch (1–1000)
committerName	string	Ever Works Agent	Commit author name
committerEmail	string	agent@ever.works	Commit author email

Sandbox Workspace

The cloud-default workspace provider. Runs plain git in an ephemeral job sandbox: every provision is a fresh shallow clone and the remote task branch is the durable identity, so a re-run fetches the pushed branch instead of re-cutting it and nothing is lost when the sandbox evaporates.

Field	Value
Plugin ID	sandbox-workspace
Category	utility
Auto Enable	Yes
System Plugin	Yes
Distribution	core
Capabilities	workspace

Settings are identical to Local Workspace (baseDir, fetchDepth, committerName, committerEmail). The checkout's origin remote is always token-free: credentials arrive per operation and are injected into the URL of that single command invocation only — never written into git config, the stamp file, or the working tree, because the checkout runs untrusted repository code. Tokens are scrubbed from every error message before it can reach a log.

PTY Local

Backs the streaming Terminal tab on an Agent's detail page. Implements terminal-stream by spawning a local pseudo-terminal and relaying it to attached viewers, with driver and viewer roles enforced by apps/api/src/terminal/terminal-attach.controller.ts (/api/agents/:id/runs/:runId/terminal). See Agent Terminals.

Field	Value
Plugin ID	pty-local
Category	utility
Auto Enable	No
Capabilities	terminal-stream

Settings:

Setting	Type	Description
defaultCols	number	Initial terminal width before the client sends a resize
defaultRows	number	Initial terminal height before the client sends a resize

node-pty is loaded as a runtime require rather than a static import, so a worker image without the native prebuild degrades instead of crashing at module load.

Browser Automation

Headless Chromium automation via Playwright, behind a default-deny navigation allowlist that is re-checked on every redirect hop. Exposes exactly four verbs so an Agent can research a page it cannot simply fetch.

Field	Value
Plugin ID	browser-automation
Category	utility
Auto Enable	No
Distribution	registry
Capabilities	browser-automation

Verb	Purpose
navigate	Go to a URL; returns the final URL, HTTP status, title and the full redirect chain
extract	Pull text / html / attribute values out of the rendered DOM by CSS selector
screenshot	Capture the viewport, the full page, or one element as base64 PNG or JPEG
act	Run a bounded ordered list of click / fill / select / press / hover / wait

Settings:

Setting	Type	Default	Description
allowedHosts	array	[]	The navigation allowlist. Empty refuses every navigation. Falls back to PLUGIN_BROWSER_AUTOMATION_ALLOWED_HOSTS (comma-separated)
timeoutMs	number	30000	Wall-clock budget for navigation and each action, clamped to 1000–120000
headless	boolean	true	Leave on for any server runtime
subresourcePolicy	string	public-only	public-only lets a page load assets from any public host while still blocking internal targets; allowlist restricts assets too
allowPrivateNetwork	boolean	false	Advanced escape hatch for internal staging hosts. Requires an explicit host list — the * entry is dropped while it is on, so it can never open the whole internal network
executablePath	string	—	Absolute path to a Chromium binary; empty uses the Playwright-managed browser
channel	string	—	Optional Playwright channel (chrome, msedge) instead of the bundled Chromium

Allowlist syntax:

Pattern	Matches
example.com	that host exactly (not its subdomains), on any port
*.example.com	any subdomain, but not the apex
example.com:8443	that host, only on port 8443
*	any public host — private and internal targets stay blocked

Anything else (a scheme, a path, an @, an embedded wildcard) is rejected as a configuration error rather than interpreted loosely. Refusals throw BrowserNavigationBlockedError with a stable code: invalid_url, scheme_blocked, credentials_in_url, private_address, not_allowlisted, dns_private_ip, or dns_lookup_failed. The SSRF guard sits underneath the allowlist, so an allowlisted hostname that resolves to 127.0.0.1 is still refused — a DNS-rebinding defence. Sessions opened with open() must be released with close(); the last close shuts the shared browser down.

How to enable and configure a plugin

1. Open Settings → Plugins (/settings/plugins) and pick the category tab — the URL is /settings/plugins/<category>, e.g. /settings/plugins/ai-provider, /settings/plugins/storage, /settings/plugins/connector.
2. Toggle the plugin on. Plugins with configurationMode: user-required stay inactive until their required settings are filled; admin-only plugins are configured by an instance administrator; hybrid plugins accept either environment variables or per-user settings.
3. Fill the settings form. Fields marked secret are encrypted at rest and rendered masked; fields with an environment-variable fallback show the resolved value when the env var is set, so you can leave them blank on a self-hosted instance.
4. Use the plugin's Test / Verify action where one exists — notification channels have Send test (POST /api/notification-channels/:id/test), and any plugin can be probed with POST /api/plugins/:pluginId/validate-connection or read back from GET /api/plugins/:pluginId/connection-status. GET /api/plugins lists every discovered plugin with its manifest and health.
5. Set a capability default where the platform needs one winner (AI provider, search, pipeline, storage, vector store, task tracker). Per-Work overrides live on the Work's own plugins page at /works/:id/plugins, backed by POST /api/works/:workId/plugins/:pluginId/capability.
6. From the CLI: ever-works plugins catalog lists what is installable, ever-works plugins install <id> pulls a distributable plugin (optionally --version and --integrity sha512-…), ever-works plugins install-status <id> reports the per-replica installer state, and ever-works plugins uninstall <id> removes it. See CLI Quickstart.

Related

• Plugin System Overview · Architecture · Settings & Configuration
• Plugin Categories & Capabilities · Creating a Plugin · Plugin API Reference
• Feature guides: Plugins · Storage Backends · Secret Stores · Job Runtimes · Connectors · Notifications · Goals
• Deployment: Kubernetes Deployment · Managed Hosting · Custom Domains