Tedix

io.github.tedix-hqv0.2.0Updated Oct 8, 2026

Open-source AI coworkers that show their work: approval before risky steps, a record of every run.

VerifiedStreamable HTTPWeb executableAI & MLSecurity & MonitoringProductivity & Workflow

Overview

AI-generated overview

Tedix provides AI coworkers that run recurring work with approval gates and a recorded audit trail of every run.

What it does
Tedix gives an assistant AI coworkers, called tedis, that take on recurring jobs with their own memory, skills and policies. Work items carry an owner, a risk level and a budget, and risky steps wait for approval before proceeding. Every run is saved as a record showing the worker, the tools it used, the approval and the result. It connects to Claude, ChatGPT Work, Codex and other MCP-compatible clients through an MCP gateway.
When to use it
Use it when you run AI agents but cannot tell whether they actually finished, when a human must approve before an agent touches money, customers or production, or when you want recurring work handled without constant supervision.
Requirements
The remote endpoint is at connect.mcp.tedix.dev; Tedix Cloud is in invited beta and access is requested through the contact page. Running it locally needs Git, Bun and Node.js 22 or later, with no account or cloud credentials required. Model turns are off by default and can be enabled through Workers AI on your own paid Cloudflare account.
Before you install
Risky steps are gated behind human approval, and nobody may approve their own request. Enabling model turns uses your own paid Cloudflare account, so inference may incur charges. Local data is stored under .wrangler/run-local. Cloud access is an invited beta, and the plugin is at version 0.1.0.

Installation

In SourceWeft

  1. Open Tedix in the dashboard and add it to a workspace.
  2. Enable the server for the chats that should use its tools.

Web executable via Streamable HTTP. Remote servers run from the web runtime once configured in a workspace.

Other MCP clients

Add this to your client's mcpServers config.

{
  "mcpServers": {
    "tedix": {
      "type": "http",
      "url": "https://connect.mcp.tedix.dev/mcp"
    }
  }
}

README

Tedix

[Tedix astronaut mascot waving next to the tagline: AI coworkers that show their work.]

[Status: beta] [License: AGPL-3.0] [GitHub stars]

Website · Docs · Discussions

Tedix gives your team AI coworkers, called tedis, for recurring work. They ask before risky steps, and every run leaves a record: what the tedi did, which tools it used, and who approved it.

If a chatbot is a smart intern you have to watch, a tedi is a teammate with a job, a budget and a paper trail.

[A supplier folder in a Tedix OS workspace]

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

How it works

StepWhat happens
1. Give it a jobAsk in Home. Tedix answers directly or hands the request to a tedi with its own memory, skills and policies.
2. It asks before risky stepsWork Items carry an owner, a risk level and a budget. Risky steps wait for approval; nobody approves their own request.
3. Open the recordEvery run is saved: the worker, the tools it used, the approval and the result.

Quick start

Run it locally

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

sh
git clone https://github.com/tedix-hq/tedix.git && cd tedixbun run-local

The launcher starts the API, database, and web app and prints Tedix OS: http://localhost:3030. Open it and name your local organization; Tedix creates the organization, your owner account, and a first tedi. Data stays in .wrangler/run-local.

Local mode makes no model calls by default. To turn on model turns through Workers AI on your own paid Cloudflare account:

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

--smoke checks onboarding and exits; --demo loads sample data. See Getting started.

Tedix Cloud (invited beta)

Ask for access through the contact page. With access:

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

Then follow Connect to Tedix Cloud.

Works with

  • Claude (chat, Cowork and Claude Code), ChatGPT Work and Codex through the Tedix plugin.
  • Other MCP-compatible clients through the MCP gateway.
  • Built on Cloudflare Workers, Durable Objects, Workflows, D1, and R2.

Who it's for

✅ You run AI agents but can't tell whether they really finished.
✅ You need a human to approve before an agent touches money, customers or production.
✅ You want recurring work done without babysitting it.

Docs and concepts

This repository is the whole product and the source Tedix Cloud is built from. Core concepts defines tedis, Home, Work Items and the MCP app platform; Release status lists what is available today.

Community, contributing and security

Ask questions and share what you build in GitHub Discussions. Issues and ideas are welcome; maintainers write the code (Contributing). 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.

Source: README.md at commit 9b15a44

Tools

0
Tool metadata has not been indexed yet.

Version history

1
  1. v0.2.0LatestOct 8, 2026