tedix

Tedix

Website · Docs · Discussions

Hand off the work. Tedix brings back only the decisions that need you, and handles the rest.

The problem

If you use AI assistants for real work, this probably sounds familiar:

• You explain your company again in every new chat, to every tool.

• An assistant says "done", and you can't see what it actually did.

• You copy results by hand from one assistant to another.

• Requests for your OK are spread across many open chats, so you keep checking on them.

• Two people, or two assistants, end up doing the same job.

What Tedix does

• One shared list of work. People and assistants work from the same list, so everyone sees who owns what and nothing is done twice.

• Context that follows you. Your company's facts, your preferences and past decisions reach whichever assistant you use: Claude, ChatGPT, Codex or Tedix's own.

• A record of every step. Each piece of work keeps what was used, what came out and who said yes.

• Decisions one card at a time. What needs you waits in one short stack, the Deck. Clear it on your phone or at your desk. Everything else goes to a timeline, the Feed, that you read when you want to.

• It stays yours. Your goals, your decisions and what your team taught the assistants belong to your organization, even when you change models or tools.

Tedix's own assistants are called tedis. A tedi has one job, its own memory and only the access you give it. It asks before anything risky, and nobody can approve their own request.

A tedi reviews one supplier folder, writes a decision note that cites its sources, and asks before sending anything. Run the small version in examples/supplier-review.

Examples of recurring jobs

• Find conversations worth joining. A tedi looks for public threads, for example on Reddit, where your product could help, and drafts a reply. You get one card per thread: post it, edit it or skip it.

• Match invoices against stock. A tedi compares supplier invoices with what actually arrived and shows you only the mismatches.

• Act on critical error reports. A tedi summarizes a new error report, finds who owns the problem and asks before alerting anyone.

Who it's for

✅ You use AI assistants every day and are tired of repeating yourself. ✅ You can't tell whether an assistant really finished the job. ✅ You want a person to say yes before an assistant touches money, customers or production. ✅ You want recurring work done without watching over it.

Try it in five minutes

On your computer, no account

You need Git, Bun, and Node.js 22+ (real Node, not a Bun shim). No account or cloud credentials are needed.

git clone https://github.com/tedix-hq/tedix.git && cd tedix
bun run-local --demo

The launcher starts the API, database, and web app and prints Tedix OS: http://localhost:3030. Open it to look around a fictional company with a sample Deck and Feed. Run bun run-local without --demo to name your own local organization instead; Tedix creates it, your owner account, and a first tedi. Data stays on your machine in .wrangler/run-local (.wrangler/run-local-demo for the demo).

Local mode makes no model calls by default, so tedis can't write anything yet. To turn on model calls through Workers AI on your own paid Cloudflare account:

bunx wrangler login && bunx wrangler whoami   # note your account id
bun run-local --inference --workers-ai-account=<account-id>

--smoke checks onboarding and exits. See Getting started for every option.

In the assistant you already use (invited beta)

Tedix Cloud is the hosted service. Ask for access through the contact page. With access:

• Claude, ChatGPT or Codex: follow Use Tedix in ChatGPT, Codex and Claude. Claude chat and Cowork need no install, only a connector URL.

• From a terminal, with the CLI:

  curl -fsSL https://downloads.tedix.dev/install.sh | sh
  tedix login

  Then follow Connect to Tedix Cloud.

Available today, and what is not

Release status has the details and the known limits.

Words we use

Core concepts has the full list.

For developers

• Tedix connects to Claude (chat, Cowork and Claude Code), ChatGPT Work and Codex through the Tedix plugin. Other clients that speak MCP, the open standard AI assistants use to call tools, connect through the MCP gateway.

• Any agent that reads Agent Skills can add Tedix's skills with npx skills add tedix-hq/tedix-plugins.

• It runs on Cloudflare Workers, Durable Objects, Workflows, D1, and R2.

• This repository is the whole product and the source Tedix Cloud is built from. The documentation home lists every guide.

• Give your coding agent AGENT-REVIEW.md to evaluate Tedix for your project in about 15 minutes and send feedback without installing the CLI.

Community, contributing and security

Tedix is built by coding agents (Codex, Claude Code, Devin and others) that Tedix itself coordinates, around the clock. Every commit names its Work Item and agent session in its trailers, so the history shows Tedix building Tedix.

Your agent can contribute too: have it open an issue with a diff against current main, or follow AGENT-REVIEW.md to send feedback in one step. Our agents re-author accepted fixes and credit you (Contributing). Ask questions and share what you build in GitHub Discussions. Report vulnerabilities through Security. See also Trademarks.

The CLI sends no telemetry; see Telemetry and network contact.

News

• 2026-10-07: The Tedix plugin restarted at version 0.1.0 and is now versioned independently of the CLI.

• 2026-10-06: The source is public on GitHub, with the first source prerelease, v0.1.0-beta.1.

License

Mostly AGPL-3.0; some packages are Apache-2.0 or MIT. See Licensing and Third-Party Notices.