Replace Confluence with a self-hosted wiki on your own domain. Let any AI agent read it — while every agent only ever sees what the person asking is allowed to see. Free to self-host. Audited. Yours.
# spin up your own Letwrites — one command, real HTTPS $ git clone https://github.com/Lynne-ang/letwrites $ cd letwrites/wiki/deploy && cp .env.example .env # set your domain $ ./deploy.sh docs.yourcompany.com is live (Let's Encrypt TLS) wiki + permission engine running — your data
It got expensive. Google Drive is cheap but loses the structure. And now you're plugging AI agents into your knowledge — which opens a risk nobody has solved.
Per-seat pricing that climbs every renewal, for features most of the team never touches.
No wiki hierarchy, weak search, no canonical "this is the doc." Knowledge gets lost.
Point an AI agent at your docs and it can surface an HR file or board deck to the wrong person.
Simple enough for your IT team to run in an afternoon. No vendor in the loop.
One command stands up Letwrites on your server with automatic HTTPS. It runs on your infrastructure — the content and the domain are yours.
One command exports your space. Text, headings, lists, code blocks, tables, images, and cross-page links convert cleanly. Anything that can't (Jira/drawio macros) is listed in a report — never silently mangled.
Admins create roles (HR, Security, Engineering) and set who can see each shelf, book, or page — down to a single doc. This one permission model is the source of truth, and it governs both people and AI agents. Set access once; it covers everything.
Connect any AI agent. On every read, Letwrites enforces the exact permissions your IT set, and writes a tamper-evident log of who (and which agent) accessed what. The agent is treated as untrusted: it physically cannot return a document the person asking isn't cleared for.
An agent answers over your wiki — but only ever from what the person asking is allowed to see.
And every access is logged, tamper-evident: verify → {"valid":true}. Edit the log to hide an access and it flips to {"valid":false}.
The core is free and yours to run — the wiki, the Confluence migration, permission-safe agent access, and an audit log. Pull the image, point a domain, and you have a production Letwrites on your servers. No account, no purchase order, no data leaving your walls.
# clone, set your domain, and bring it up $ git clone https://github.com/Lynne-ang/letwrites && cd letwrites/wiki/deploy $ docker compose up -d wiki + permission engine + database, behind auto-HTTPS
Need SSO, the governance dashboard, more connectors, or managed hosting? Those are paid add-ons — but the core stays free and your data stays on your servers. See pricing →
Confluence → your wiki: text, code, images, tables, hierarchy, links. A report lists exactly what needs a human.
Agents enforce your existing permissions on every read. Treated as untrusted — can't be tricked into leaking.
Hash-chained log of every allowed and denied access. Edit it and verification breaks — provably.
Runs on your servers, your domain. Content never touches a vendor's cloud. The whole point.
Works with Claude, Copilot, Cursor, internal bots — any agent. Not locked to one AI vendor.
Roles and per-page permissions, managed in one place, governing both humans and AI.
Open-core: the core is free forever. SSO and add-ons unlock on a subscription — and even then, you still self-host and your data still stays on your servers.
Prices are illustrative and billed per active user, annually. Community edition is free under Apache-2.0, forever — no seat limits.
See the migration. Watch an agent get blocked from a doc it shouldn't see. Read the audit log. Then decide. It's free to try on your own server.