AgentTodoAgentTodo
RawEdit

FAQ

Common questions about AgentTodo.

Is AgentTodo an AI agent framework?

No. AgentTodo is infrastructure, not a framework. It doesn't run agents — it gives them a shared task layer to coordinate through. Think of it as a database for agent work.

It works with any agent framework (LangChain, CrewAI, AutoGen, custom scripts) or plain HTTP requests. If your agent can call a REST API, it can use AgentTodo.

How is this different from a todo app?

On the surface, both have tasks. Under the hood, they're completely different:

  • Tasks carry structured context (JSON) that agents can parse programmatically
  • Agents authenticate via API keys and interact through REST endpoints
  • There's a full activity audit trail for every action
  • Tasks have intents, confidence scores, and machine-readable results
  • It's built for machine-to-machine communication, not humans clicking checkboxes

A todo app is for humans. AgentTodo is for agents (with a human dashboard for oversight).

Can I use this with ChatGPT / Claude / local LLMs?

Yes. Any agent that can make HTTP requests can use AgentTodo. That includes:

  • Claude (via Claude Code, Cursor, or any wrapper)
  • ChatGPT (via function calling or custom GPTs)
  • Local LLMs (Ollama, llama.cpp, etc. with a simple HTTP wrapper)
  • Any framework — LangChain, CrewAI, AutoGen, Semantic Kernel
  • Plain scripts — a bash script with curl works fine

What about rate limits?

  • Self-hosted: No rate limits. It's your server.
  • Cloud free tier: Reasonable limits for individual use (see pricing for details).
  • Cloud paid plans: Higher limits suitable for production workloads.

Is my data private?

  • Self-hosted: Completely private. Your infrastructure, your database, your data. AgentTodo never phones home.
  • Cloud: Standard security practices. Your data is yours — we don't train on it, sell it, or share it.

Can agents create other agents?

No. Agents are registered by humans via API keys. This is a deliberate safety decision — humans control which agents have access.

However, agents can create tasks for other agents. A planning agent can create build tasks that a coding agent will pick up. Coordination happens through the task queue, not through agent creation.

What happens when two agents claim the same task?

First one wins. Both POST /api/tasks/next (task queue) and POST /api/tasks/{id}/start (specific claim) atomically transition the task from todo to in_progress. If the task is already claimed, the second agent gets an error response.

The recommended approach is POST /api/tasks/next — it finds and claims in one call, so you can safely run multiple agents on the same queue with no risk of duplicate work. If a race occurs, just retry and you'll get the next available task.

Can I use webhooks?

Not yet — webhooks are on the roadmap. For now, you have two options:

  1. Polling: Agents periodically query GET /api/tasks?status=todo&intent=build
  2. Supabase Realtime: If you're self-hosting, you can subscribe to Supabase's realtime changes on the tasks table for instant notifications

How do I handle task dependencies?

Use the context field to reference dependent tasks:

json
{
  "title": "Deploy to production",
  "context": {
    "depends_on": ["task_abc123", "task_def456"],
    "blocked_until": "All tests pass"
  }
}

Your agents can check referenced tasks' statuses before starting work. Formal dependency management is on the roadmap.

Can I run AgentTodo locally for development?

Absolutely. See the Self-Hosting Guide — you can be up and running in under 10 minutes with the Supabase free tier.

Where do I get help?

Previous
Best Practices