Todos

Todos is a task-driven workspace for humans and AI agents. You give the "Chief" agent a one-sentence goal; it splits the work into standalone todos, assigns each by trade (frontend, backend, QA, DevOps), and starts them immediately — then keeps tracking progress and reports back as each stage moves forward. It is positioned as a lightweight, minimal, and restrained agent tool that works particularly well with DeepSeek models.

Key Features

• Chief Agent Orchestration: A coordinator agent that decomposes goals, assigns tasks by role, monitors progress, and reports autonomously.
• 700+ Models: GPT, Claude, DeepSeek, local models via Ollama, or any OpenAI-compatible endpoint — each agent picks its own model and thinking level.
• Runs on Your Machines: Execution happens on executors you own (Node.js and git only, no GPU needed); builds run in isolated git worktrees and PRs go straight to GitHub.
• Plan-then-Build: Every todo produces an approach before implementation, with a separate reviewer agent option and the diff, plan, and test runs kept on one thread.
• Live Preview & Download: Browse, preview, and download every file an agent delivers, without deploying.
• Cost Transparency: Itemized token usage and cost per model, with no markup on inference — your keys go straight to the providers.
• Skills & Memory: Learnings are banked as skills attached to agents and reused automatically; conclusions written to agent memory across rounds.

Use Cases

Who Should Use This Tool?

• Indie Developers: Solo builders who want an agent team (frontend, backend, QA, DevOps) working on their repo without cloud-sandbox lock-in.
• Small Teams: Teams of up to 3 people that want a shared Chief agent routing work across projects.
• DeepSeek Users: Developers who want a lightweight agent tool optimized for DeepSeek model economics with cloud coordination.

Problems It Solves

1. Agent coordination overhead: Instead of prompting one agent to do everything, Chief divides work by trade and runs tasks in parallel.
2. Vendor sandbox lock-in: Execution stays on your hardware, so speed and parallelism are yours to scale; the cloud only coordinates.
3. Cost opacity: Per-model token and cost accounting with no inference markup keeps budgets predictable.

Pricing

You pay your model providers directly with the API keys and subscriptions you already have — including ChatGPT/Codex or GitHub Copilot subscription sign-in without an API key.

Advantages & Unique Selling Points

Compared to Competitors:

1. Human + agent teams in one workspace: Mix human members and agent members; Chief routes work across both.
2. Local execution, cloud coordination: Not a vendor sandbox — your machines run the builds, so there is no per-seat runtime fee.
3. Model-neutral economics: 700+ models with per-agent selection and zero inference markup.

What Makes It Stand Out:

• The "Chief" abstraction that makes a multi-agent team feel like managing a single assistant.
• Extremely lightweight and restrained design that pairs well with DeepSeek's prefix-cache pricing.
• Two-stage plan/build with independent review built into every todo.

Getting Started

Quick Start Guide

1. Install the app: Visit todos.dev and install the PWA on any device.
2. Create a team: Invite human members by email and add agents with their model assignments.
3. Register an executor: Run tds start on a machine you own to register it as an executor.
4. Give Chief a goal: One sentence, and Chief splits, assigns, tracks, and reports.

Integration

Integrates with:

• GitHub (PRs), any OpenAI-compatible model endpoint, Ollama local models
• ChatGPT/Codex and GitHub Copilot subscriptions (one-time sign-in)
• Agent skills installed from GitHub

Frequently Asked Questions

Does it run my code in the cloud?

No — execution happens on machines you own. The Todos server only coordinates projects, conversations, and reviews.

Does it work well with DeepSeek?

Yes. Its lightweight, restrained design is particularly suited to DeepSeek models, and it supports any OpenAI-compatible endpoint.

What are the requirements for an executor machine?

No GPU required. Any computer that runs Node.js and git can be an executor; local models via Ollama are optional.

Alternatives

If Todos isn't the right fit, consider these alternatives:

• Reasonix: DeepSeek-native terminal agent in a single Go binary.
• CrewAI: Role-based multi-agent orchestration framework for developers.
• LangGraph: Low-level stateful agent orchestration from LangChain.

Tips & Best Practices

1. Start with a clear one-sentence goal: Chief's decomposition quality depends on how concrete the goal is.
2. Assign models deliberately: Put high-capability models on demanding roles and light models on simple work.
3. Teach routines once: Use skills to encode repeatable processes (release runbooks, triage playbooks) so agents reuse them.

Conclusion

Todos rethinks agent teams as a task-driven workspace where a Chief agent decomposes, assigns, tracks, and reports — with execution staying on your own hardware and no inference markup. If you want a lightweight, cost-transparent multi-agent setup that works especially well with DeepSeek, Todos is worth trying.