Providers

TatsuCode supports multiple provider types so you can choose the best tradeoff between cost, speed, privacy, and model variety.

Independent product. TatsuCode is not affiliated with, endorsed by, or sponsored by Anthropic, OpenAI, Google, GitHub/Microsoft, OpenRouter, or any other provider listed below. All product names and trademarks belong to their respective owners. See Third-Party Services for full attributions.

Primary command: /connect Alias: /providers

See also: CLI vs API/Subscription Providers — when to use which, what each can and cannot do as a Task Agent or SubAgent, and how Cross-Agent Relay fits in.

Provider Options at a Glance

OAuth Providers (Use Existing Subscriptions)

If you already pay for AI subscriptions, connect supported routes directly and use that quota inside TatsuCode.

OpenAI Plus / Pro

Subscription: chatgpt.com

To connect:

1. Run /connect
2. Select OpenAI Plus/Pro
3. Sign in and approve access
4. Complete sign-in and return to TatsuCode when prompted

For broader model coverage, use OpenRouter API key access.

Claude Agent (Claude Code)

Subscription: Claude Pro, Max, Team, or Enterprise on claude.ai

To connect:

1. Sign in to the Claude Code CLI first — run claude and complete its own /login flow with your Anthropic account. TatsuCode does not handle this step (it stays out of the Anthropic login to respect Anthropic's terms of service); the CLI stores your token under ~/.claude/.
2. In TatsuCode, run /connect
3. Select Claude Agent (Claude Code) to enable it
4. The Claude Code models appear in /models

Claude Code runs as a CLI provider — it brings its own auth and tool loop, and produces archived report.md artifacts on every run. For deep details, see Claude Code and the CLI Providers overview.

Free Claude.ai accounts can't run Claude Code; an active paid subscription is required.

GitHub Copilot (removed)

The GitHub Copilot connection, switched off since June 4, 2026, is removed in TatsuCode 1.2:

• Leftover Copilot models disappear from your model list, and a session that used one reopens on another connected model.
• To keep using the same models (Claude, GPT, Gemini, and others), connect OpenRouter.
• To also withdraw the access you granted on GitHub, open GitHub's Settings → Applications → Authorized GitHub Apps and revoke the Copilot entry. TatsuCode signed in through the same GitHub Copilot entry some editors' Copilot extensions use, so revoking it can sign those out too.

OpenRouter (API Key)

OpenRouter gives you access to a large cross-provider catalog from one key.

Why OpenRouter

• One API key for many model families
• Fast switching between providers
• Pay only for usage
• Easy fallback when one provider is unavailable

Setup

1. Create account: openrouter.ai
2. Add credits
3. Create API key: openrouter.ai/keys
4. Run /connect
5. Open OpenRouter
6. Paste key and save

Quota and Spend Monitoring

Use:

/usage
/usage-quota

/usage-quota now includes clearer subscription/quota visibility and OpenRouter balance/usage summaries in recent builds.

Local Providers (llama.cpp recommended)

For local, fully-offline, and privacy-first workflows, TatsuCode connects to any local OpenAI-compatible endpoint. llama.cpp is the recommended runtime — it's the lightest, most portable, and most direct way to serve a GGUF model to TatsuCode, and it runs anywhere from a laptop to an air-gapped workstation with no extra layers in between.

Recommended: llama.cpp

1. Build or download llama.cpp and grab a GGUF model (Gemma, Qwen, Kimi, DeepSeek, or any GGUF you already have)
2. Start its server, e.g. llama-server -m your-model.gguf --port 8080
3. In TatsuCode, run /connect → Add Custom Provider
4. Enter a name and the server URL (e.g. http://localhost:8080/v1)
5. Save and test, then run /models to pick the model

Alternative local runtimes

If you'd rather use a managed app, Ollama and LM Studio both expose an OpenAI-compatible endpoint and connect the same way — add them as a custom provider using the URL each one prints when it starts. Any other OpenAI-compatible local server works too.

Setup Flow (any local server)

1. Start your local server (llama.cpp, Ollama, LM Studio, or another OpenAI-compatible endpoint)
2. Run /connect
3. Choose Add Custom Provider
4. Enter name + URL
5. Save and test
6. Run /models to pick a model

Decision Models (TypeSafe AI)

A decision model is not a chat model. TypeSafe AI's System One model answers typed questions about a piece of state (choose one option, place on a scale, yes or no) with probabilities and confidence, and it does it in roughly 100 to 500 ms. The agent uses it for bounded judgments inside a task and can build it into your own code.

Setup inside /connect:

1. In the Decision Models row, click Configure next to TypeSafe AI.
2. Under Access, pick TypeSafe API key and paste a key from console.typesafe.ai, or pick OpenRouter to reuse the OpenRouter key you already connected (decisions run as ~typesafe/jev-latest and appear on your OpenRouter bill).
3. Click Save & Verify. A TypeSafe key is checked against TypeSafe's API and stored locally with your other credentials; the OpenRouter route is confirmed with one small decision.
4. Load the typesafe-ai skill with /skills when you want the agent to use it.

The key is yours; requests go from your machine to TypeSafe, or to OpenRouter when you choose that route, under their terms. See Decision Models for what the model can and cannot do.

Web Search Providers

TatsuCode can be configured with your own search API keys.

• Google Programmable Search
• Brave Search API

Configure them inside /connect under the Web Search section.

Connection Management Tips

Inside Provider Connections you can:

• Connect/disconnect providers
• Verify API keys
• View available model counts
• Switch fallback providers if one is down

If a provider is removed, TatsuCode now handles model fallback more gracefully so you can keep working.

Troubleshooting

Invalid API Key

• Re-copy key from provider dashboard
• Remove leading/trailing spaces
• Confirm key is active

OAuth Won't Complete

• Try disconnect + reconnect
• Retry promptly if the sign-in step expires
• Retry from a fresh browser session

Local Provider Not Reachable

• Confirm your local server is running
• Re-check the configured endpoint
• Validate local network/firewall policy

Next Steps

• Models — choose the best model per task
• Settings — tune behavior and defaults
• Troubleshooting — common issues and fixes