Onyx Agent Wiki
https://github.com/onyx-dot-app/agent-wiki- Category
- Productivity
- Rank
- No. 837Tools index
- Pricing
- Open Source
- Type
- TOOL
- Builder
- onyx-dot-app
- GitHub
- 37 stars
- Latest release
- agent-workspace-0.3.35
- Date
About
Self-updating knowledge base wiki app to supercharge human and agent collaboration.
What it does
Onyx Agent Wiki stores Markdown pages in a Git-backed hierarchy while keeping operational data in Postgres. People edit through a collaborative rich-text interface. Agents read and write through MCP, while external systems submit documents through an API. Background jobs propose cleanup, triggers evaluate changes, and inherited policies control which content automation may rewrite.
Why it's ranked here
The design connects durable documentation, agent access, and human governance unusually well. Git preserves page history, shared access rules constrain both people and agents, and approval-based organization limits unattended cleanup. It also exposes serious operating concerns clearly, including split-state backups, search rebuilding, secret management, queues, metrics, and deployment options.
What's good
The permission model applies the same readable scope to agents, searches, and human users. Folder grants and update policies inherit while allowing narrower overrides. Collaborative editing includes automatic merging, participant cursors, media, threaded comments, and recoverable deletion. Natural-language triggers can log events, call APIs, send webhooks, or start downstream work. Prometheus instrumentation covers requests, ingestion, search coverage, media, and queue health.
Tradeoffs
This is a service stack, not a lightweight folder viewer. Complete recovery requires matching backups of the Git repository and Postgres database, plus preservation of the original encryption secrets. OpenSearch and Redis add operational components, although their state is rebuildable or transient. Automated ingestion depends on model configuration and tuning. Email requires an external SMTP account. The local coding-agent launcher supports macOS only.
How to use it well
Use it when teams want agents to maintain project documentation without bypassing human permissions, editorial rules, or change history. Start with approval-required organization proposals and conservative inherited update policies. Connect task agents through MCP, route source systems through the API, and reserve triggers for consequential changes. It does not replace object-storage retention, encryption policy, SMTP delivery, or broader infrastructure backup planning.
Technical notes+
The FastAPI composition root in backend/app/main.py mounts API surfaces for wiki content, MCP, co-editing, comments, permissions, triggers, media, sessions, webhooks, authentication, and metrics. Runtime settings and production secret checks live in backend/app/config.py. frontend/next.config.js builds a standalone Next.js frontend and proxies development API traffic, while frontend/vitest.config.ts configures DOM-backed TypeScript tests. packages/agentwiki-launcher/src/cli.ts and packages/agentwiki-launcher-go/cmd/agentwiki-launcher/main.go exchange launch codes, materialize MCP configuration, validate or confirm pinned endpoints, and open Claude or Codex sessions. docs/backups.md documents paired Git bundle and Postgres dump recovery; docs/AGENT_WIKI_MARKDOWN_STANDARD.md defines the supported Markdown dialect and excludes live raw HTML.
Observed
- Languages
- Python backend, TypeScript frontend and launcher, plus a Go launcher implementation.
- Install surface
- Shell installer and Docker Compose deployment are documented.
- Cluster deployment
- Kubernetes deployment supports Helm, with optional Terraform for EKS.
- Interfaces
- Browser editor, HTTP API, MCP server, webhooks, event polling, and local launcher CLI.
- Storage model
- Markdown pages and history live in Git; users, policies, permissions, comments, triggers, queues, settings, and media live in Postgres.
- Platform support
- The Go local-agent launcher installs its application and URL handler only on macOS.
- Testing structure
- The frontend includes Vitest configuration using happy-dom, and the backend entry point documents TestClient-based tests.
Read from README.md, packages/agentwiki-launcher/src/cli.ts, packages/agentwiki-launcher-go/cmd/agentwiki-launcher/main.go, docs/backups.md, docs/email-setup.md, docs/AGENT_WIKI_MARKDOWN_STANDARD.md, frontend/next-env.d.ts, frontend/next.config.js, frontend/vitest.config.ts, backend/app/main.py, backend/app/config.py, backend/evals/_cli.py, backend/app/metrics.py, backend/evals/schema.py.
What it can do
Create and maintain a self-updating knowledge base
Information from various sources → Organized wiki with automatically updated content
Enable collaborative editing between humans and AI agents
Content contributions from both humans and agents → Unified wiki pages with mixed human-AI content
Automatically update wiki content
New information or data changes → Refreshed wiki pages with current information
Track collaboration between humans and agents
User interactions and agent activities → Collaboration history and activity logs
Organize knowledge into structured wiki format
Raw information and data → Structured wiki pages with proper formatting
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.