The answer is hidden somewhere across your sources. passageway finds it.
passageway searches all your sources at once, wiki, notes and documentation, and gives your agent exactly the passages that answer the question, each linked to its source.
- On premise. Your knowledge stays on your infrastructure.
- Connected. Confluence, Trilium, Markdown and Wikipedia work out of the box.
- Open. A documented protocol lets you add any source or embedding model, in any language.
Free for private use. Companies need a licence (details); trying it at work for fewer than 32 days is covered by the PolyForm Free Trial licence (LICENSE-FREE-TRIAL).
It serves the passages as one recall tool over MCP, stays silent when nothing is relevant enough, and does no generation: the calling agent reads the passages. An eval harness (passageway eval) measures retrieval quality on your own questions, so you can publish the numbers.
- Install and run and configuration
- How it works
- What passageway stands for: the yardstick for every change
- Run it as a local service or as the container
ghcr.io/getpassageway/passageway; in Claude Code, the plugin installs and connects it for you and brings the prompt hook - Write a connector, the connector protocol and the Python SDK that does its plumbing
- Bundled connectors: Wikipedia, Markdown, llms.txt, MkDocs, Sitemap, Trilium, Confluence, GitHub; bundled embedder: OpenAI-compatible
- For AI agents: skills and
llms-full.txt; in Claude Code,/plugin marketplace add getpassageway/passagewayand/plugin install passageway@passagewaygive you the plugin, then/passageway-installsets everything up
Documentation: https://getpassageway.com (the pages are in docs/).
Status: early. The engine started as the recall engine of trilium-recall and is being generalized. The code is source-available under the PolyForm Noncommercial License 1.0.0 (free for private use; companies need a licence), and the SDK is MIT; the Wikipedia articles of the sample corpus in corpus/ are CC BY-SA 4.0 (see corpus/LICENSE.md).