Loading…
ADA
Cookie notice
We use first-party cookies to keep you signed in, protect your account, and understand how the site is used so we can improve it. Read our Privacy Policy for details.
By continuing you agree to our Terms and Privacy Policy.
ADA plans, builds, tests, and reviews code across your repo. Native desktop. Real tools. Real review loops. Your account, credits, and subscription live here.
Free tier — no credit card required. $1 in credits to try the agent.
Bring your own model — works with
Built for shipping
ADA is not a chat box bolted on top of completions. It is a multi-agent system with explicit plans, safe execution, and a review loop you can audit at every step.
A planner agent decomposes your goal into a graph of steps. You see and approve every move before it runs.
Shell, file edits, and web tools run behind explicit approvals. Nothing escapes the workspace without your nod.
Builder, Reviewer, and Debugger agents trade work in real time. Lint and tests run on every iteration.
OpenRouter, Vortex, or direct provider keys. Route premium and cheap models per task.
Long-term project memory survives across sessions so context never resets back to zero.
Credits, subscription, invoices, and account security live in this web app — managed by Stripe and Clerk.
How it works
Tell ADA what you want — a feature, a refactor, a bug fix. No prompt engineering required.
The planner returns a step graph. Edit, prune, or send it through.
Builder makes edits, Reviewer runs lint and tests, Debugger fixes regressions. You stay in the loop.
Frequently asked
ADA is an autonomous AI software-engineering platform. Instead of giving you completions, it plans the work, edits files, runs tests, and reviews its own output across multiple coordinated agents — all from the desktop app or your browser.
Copilot and Cursor are great inline assistants. ADA runs a full multi-agent orchestrator: a Planner produces an explicit step graph, a Builder executes edits and shell commands, a Reviewer enforces lint and tests, and a Debugger catches regressions. You approve every tool call, watch every step stream in real time, and the agent can ship features end-to-end while you watch.
No. ADA ships with built-in access to OpenRouter (300+ models including Claude, GPT-5, Gemini, Llama) and Vortex AI. You can also plug in your own keys for Anthropic, OpenAI, or Google AI if you prefer. Routing happens per task so cheap models do cheap work.
Yes. The desktop app runs on your machine and only the prompt + selected context leaves your workspace. We do not train on your code. We do not store your code. Tool execution is sandboxed and every shell command requires explicit approval unless you whitelist lint / test runners.
Free to start with $1 of credits. Starter is a one-time $20 in prepaid credits. ADA Pro is $50/month and includes $40/mo in agent credits, then pay-as-you-go, plus premium model access and priority routing. Every plan is self-serve via Stripe — no sales calls.
A native Tauri app for Windows, macOS, and Linux. It speaks WebSocket to the orchestrator, owns your local workspace, and runs file edits + tool calls locally so latency stays low and your code never round-trips through a third party.
Yes. ADA reads your repository, builds a memory of your codebase and conventions, and respects your linting + test commands. It works equally well on Python, TypeScript, Rust, Go, Java, and most other languages.
Free to start. Upgrade when you want more capacity, premium models, and priority routing.