OpenWiki

There's a quiet but telling trend in dev tooling right now: nobody wants to write documentation, but everyone increasingly needs it fresh enough for an AI agent to trust. OpenWiki is the latest entrant chasing that problem, and it's unusually specific about who it's for. This isn't a wiki for humans to browse leisurely. It's a wiki for agents to query, built on Google's Open Knowledge Format v0.1, with YAML front matter, reserved index files, and Markdown links standing in for relationships between concepts.

The tool ships in two modes. Code mode generates and maintains repository documentation inside an openwiki/ folder, and it does something clever on top: it maintains AGENTS.md and CLAUDE.md files at the repo root, quietly inserting a managed block that tells coding agents to check the wiki before searching for context elsewhere. Personal mode is the more ambitious sibling — a local 'brain wiki' at ~/.openwiki/wiki that ingests from Gmail, Notion, X/Twitter, Hacker News, and web search via Tavily, storing everything as raw manifests before synthesizing it into wiki pages. You can even run multiple instances of the same connector, so one Tavily search feed tracks AI research while another tracks NBA news, addressed individually as web-search-1 and web-search-2.

What's notable is how seriously the project treats CI and automation over one-off generation. There are ready-made workflows for GitHub Actions, GitLab CI, and Bitbucket Pipelines, and the update command is smart enough to bootstrap docs from scratch if none exist yet, provided you've set the right provider and model environment variables. Scheduled runs phone home with anonymous reliability telemetry by default, though it's a one-line environment variable to opt out. On macOS, personal-mode connector schedules even get installed as LaunchAgents, which is a small detail that suggests real thought went into making this run unattended for months, not just as a demo.

Model support is broad to the point of being almost provider-agnostic. OpenAI, Anthropic, Gemini, AWS Bedrock, OpenRouter, Fireworks, Baseten, NVIDIA NIM, and a generic OpenAI-compatible slot that happily points at Ollama, LM Studio, or a self-hosted LiteLLM gateway. Bedrock users get IAM-based auth instead of a vendor key, and there's a ChatGPT-login mode that draws down your Plus or Pro plan's Codex usage instead of billing per token through the API — a nice touch for people who'd rather not manage two separate OpenAI bills.

The install story has one rough edge worth flagging: Bun users installing globally can silently trigger a native compile of the better-sqlite3 checkpointing dependency, and because Bun skips lifecycle scripts, you won't get a warning before Visual Studio Build Tools become a requirement on Windows. npm or pnpm sidestep that entirely, so the project's own docs steer people there first.