macOS · Windows · Linux

Your AI development team,
on your own machine.

Devpal runs the models you choose — Claude and Codex, API providers, or models on your own hardware — as one agent in a chat, as a team with a Lead and a Verifier, or as a venture that keeps building while you are away. Your models, your code, your computer.

See how it works

The signed desktop apps are on their way.

macOS Windows Linux

Devpal in Single mode: a chat where the agent found a double-charge bug, fixed it with an idempotency key and reports four passing tests.

Three ways to work

Start with one agent. Hand over a whole milestone when you are ready.

Single

One agent, in a chat next to your code

Ask in plain words. The agent reads the repository, edits, runs the tests and tells you what changed. Steer it while it works, attach screenshots, switch models per message.

An IDE beside the chat: files, git, language servers, debugger
Permissions you choose, checkpoints you can roll back to
Debate mode: several models argue one problem, a judge decides

Single mode: the chat view with the agent's steps, the fix and the composer.

Team

A room of agents with a Lead — and a Verifier who says no

Give the Lead a goal. It plans a task board, the members build in parallel, and every task carries acceptance criteria. Nothing is “done” until the check command passes and the Verifier has read the evidence.

Mention a member with @Name, or talk to the Lead
Pause, reopen and resume — the transcript is the record
When one subscription runs out, members move to the next

Team mode: a room with a Lead, Backend, Frontend, Tester and Verifier, a four-task board, and the Verifier passing a task with evidence.

Ventures

An idea in, milestones out — around the clock

A venture keeps a project manager, an assistant and a support desk running. It plans milestones, staffs a team room for each one and commits the accepted work. It asks you only what it cannot decide.

Backlog, owner requests and an activity log in one place
Talk to it from Telegram: status, answers, pause and resume
Credentials stay in your secret vault; the team only sees their names

Ventures mode: four milestones, the first done, the second active, two planned.

What you get

Built for real repositories, not demos.

Local first

Devpal runs on your own computer. Your code stays on your disk and goes only where your agents send it.

Any model, local ones too

Claude Code and Codex with the logins you already have; Anthropic, OpenAI, Gemini, OpenRouter, Mistral, DeepSeek, Groq or Together with an API key; Ollama and LM Studio on your own machine; or any OpenAI-compatible endpoint.

Proof, not promises

Acceptance criteria, a check command and a separate Verifier stand between “done” and done.

Connectors

Jira, GitHub, GitLab, Confluence, Figma and any MCP server. Tokens live in a secret vault, never in the database.

Memory that carries over

Lessons learned in one project are recalled in the next: gotchas, fixes, conventions.

Evals on your own code

Benchmark models on tasks from your repositories, graded by the tests and a judge.

Pricing

Every feature in every plan. You pay for seats, not features.

Personal

Pricing to be announced

One person, on their own computer.

The desktop app for macOS, Windows and Linux
Single, Team and Ventures
Unlimited projects and runs

Business