Agent-native wiki · semantic search · in Claude & ChatGPT
Connect tela to Claude and ChatGPT. The assistant searches your docs by meaning, answers with the source, and edits pages when you ask. Your team gets a fast wiki with real-time editing, semantic search, and content stored as plain markdown.
Free to start · export your markdown anytime · self-host if you prefer
When a service degrades, the on-call engineer owns the timeline until handoff. Keep this page open during an incident.
tela incident open "api-latency" --sev 2
The agent layer
Your AI already lives in the chat apps you work in, so tela connects to them — instead of bolting a chat box onto a wiki you'd have to go open. Connect once and the assistant searches your docs, reads and writes pages, and adds comments, limited to what your account can access. Below: a question answered from the wiki, then a page edited — without leaving the chat.
When a service degrades, the on-call engineer owns the timeline until handoff. Keep this page open during an incident.
tela incident open "api-latency" --sev 2
In Claude or ChatGPT, add tela as a connector.
tela.cagdas.io/api/mcp One sign-in with your tela account. No token to paste.
OAuth 2.1 The assistant searches and edits the wiki, within your permissions.
20 tools Available in the Claude and ChatGPT connector directories. Nothing to install.
Using Claude Code, Cursor, or your own client? Connect to the same address: tela.cagdas.io/api/mcp.
Give an agent a read-only key, or one that can write. A key can be pinned to a single space.
semantic_searchsearchread_chunkget_pagelist_pageslist_backlinkscreate_pageupdate_pageadd_commentmove_pagecreate_spaceupdate_spacemanage keysThe assistant reads from the wiki and writes back to it, so context carries from one session to the next instead of being pasted in each time.
See the full tool catalog →Retrieval
tela searches by meaning, not just keywords. A search for “rollback steps” returns the runbook section that never uses the word. Keyword and semantic results are ranked together, in the command palette for your team and through the connector for your agent.
Press ⌘K to search every space. Results are sections of pages, ranked by relevance, with the matching text highlighted.
Claude and ChatGPT search the same way and get the matching section back with a link to the page, so answers cite a source.
Semantic search is on by default on the hosted instance. When self-hosting, keyword search works out of the box; semantic search needs an embedding model you point tela at.
A real wiki first
Block editing, real-time collaboration, comments, version history, sharing — tela works as a wiki without any of the AI. The Claude and ChatGPT connector is one more way in, not the whole product.
Editor
Block-based editing: drag a block to reorder it, type / to insert headings, lists, tables, callouts, diagrams, and math. Every page is stored as plain markdown you can export at any time. No proprietary format.
Features
Far more than text — and all of it the day-to-day stuff your team actually uses.
tela incident open --sev 2Callouts, toggles, tables, kanban boards, code, math, Mermaid and Excalidraw diagrams, task lists — inserted from a / menu and stored as plain markdown.
Ship checklist
Tag the release, then push the image and re-probeSam
Verify /api/version matches HEAD
Purge the edge cache lastLee
Roll back in one click if it drifts
Several people edit a page at once with live cursors and no save conflicts. Comments stay attached to the text they mark.
Share any page by link, with an optional password and expiry date.
# Incident response
When a service degrades, the
**on-call engineer** owns it.
> [!note] Page @on-call first.
`tela incident open`Every page is a markdown file. Search, diff, and export them; nothing is locked in a proprietary format.
Every change is saved. Compare versions and restore any of them.
Escalate via the [[On-call rotation]] before paging anyone.
Link pages with [[…]]. Every page shows what links back to it.
Backlinks exist on every page today. The graph that draws them is on the roadmap.
Scaffold runbooks, RFCs, and postmortems in a click. On the roadmap.
Import a folder of markdown in one step, and export everything at any time.
Compare
Where tela differs from each — and the one thing each still does better.
| Versus | Where tela wins | What it still does better |
|---|---|---|
| Notion | Plain markdown you own, semantic retrieval, and a real MCP connector your agent drives from Claude or ChatGPT — not a closed block store with a chat sidebar. | Databases, templates, and all-round polish are years ahead. Want a relational workspace, not a wiki? Use Notion. |
| Confluence | Agents search and write it by meaning, your content is portable markdown, and you can run it yourself — no enterprise install, no per-seat pricing. | Jira integration, granular permissions, and governance at thousand-user scale. Atlassian shops have reasons to stay. |
| Obsidian | Built for a team: live multiplayer, a shared server with SSO and roles, and an agent API — not a single-player vault you sync by hand. | Local-first single-user is its whole point; the plugin ecosystem and graph view are unmatched. Solo? Hard to beat. |
| A git repo of markdown | Semantic + full-text search, block editing, comments, sharing, and an agent connector — over the same markdown, no PR to read a doc. | Pure version control, free, zero infra. If you just want files in a repo, you already have one. |
| Notion AI / “AI” wikis | Retrieval your own agent drives from Claude or ChatGPT, over markdown you can export — not a built-in chatbot you can only use in their app. | One-vendor convenience and a polished in-app assistant, with no embedder to think about. If you only want their chatbot, it’s simpler. |
Only tela combines all of it: semantic search your agent uses from Claude and ChatGPT, plain markdown you own, real-time editing, and SSO with roles and audit logging.
Security & team
Every space can be shared with your whole organization, opened to a specific team, or kept private to you. People get owner, editor, or viewer access — and an agent connected through Claude or ChatGPT can only ever see what your own account can.
Share a space with your whole organization, limit it to one group, or keep it private to you.
An agent in Claude or ChatGPT only ever sees what your account can. It can’t reach a space you’re not in.
Give each person owner, editor, or viewer access. Access can be added but never silently lowered.
Sign in with the SSO your team already uses. Accounts are email-verified.
For code agents, each key is read-only or read-write, can expire, and can be pinned to one space.
Public links can require a password and an expiry date. Every key request is logged.
tela is open-source and runs on your own server with Docker Compose — your Postgres, your disk, your data. Self-hosting is optional; the hosted instance is ready to use.
Read the docs →# your server, your data
cp deploy/.env.example deploy/.env
docker compose upHonest caveats: orgs are admin-provisioned (not open self-service signup) and social login isn't wired yet — by design for now. The access model is documented and the auth code is open.
Credibility
Backend, frontend, and the MCP server are open. See exactly what runs.
GitHub →tela.cagdas.io runs the same code. Use it before you commit.
Open it →A real MCP connector, submitted to the Claude and ChatGPT directories and versioned against the backend.
How it works →tela is at v0 and usable today. No fabricated logos, no "trusted by thousands" — just the code, a running instance, a connector you can add, and a spec you can read.
Questions
Yes. tela runs a remote MCP server with OAuth sign-in; add it as a connector in Claude or ChatGPT (it’s submitted to both directories) and your agent searches, reads, and writes your wiki — scoped to your account. Code agents like Claude Code or Cursor point at the same URL with a token.
tela ranks results by meaning as well as by keyword, so a search returns the section that answers your question even when it doesn’t contain the exact words. Your team searches from the command palette; Claude and ChatGPT search the same way through the connector and get a link back to the source.
Only if you self-host, and only for meaning-based search — you point tela at an embedding model (the hosted instance is already set up). Keyword search always works on its own.
Yes. The editor is full block editing — drag, slash menu, turn-into, tables, diagrams — but pages.body is plain markdown. There is no block table; reordering a block reorders markdown lines. Import a directory, export anytime.
Through the Claude and ChatGPT connectors they sign in once with OAuth — no token to handle. For code agents like Claude Code or Cursor, you create an access key that’s scoped to read or write, can expire, and can be limited to a single space.
Yes — SSO, organizations and groups, and per-space roles (owner / editor / viewer) with hard invariants. Keys are scoped and audited. The access model is documented and open.
Yes. It’s open and self-hostable with Docker Compose (Postgres, plus an optional embedder for semantic search). Your data on your disk, your markdown exportable. Self-host is the option, not the requirement — the hosted instance is ready to use now.
Start free on the hosted instance, or self-host it yourself.
Free to start · markdown you own · self-host whenever you want