claudio-cli

🧪 Open Beta. Claudio is in active open beta — expect frequent updates, rough edges, and fast iteration on feedback. We're shipping improvements daily.

Bring your own model. Run real pipelines.

Claudio is a plug-and-play AI dev platform that runs on top of the model provider YOU pick — no per-token API bill eating your budget while agents think. Use your existing Claude Code subscription, sign in with your ChatGPT Plus / Pro / Team account, paste an OpenRouter / Anthropic / OpenAI / DeepSeek API key, or any of the other supported providers. Claudio wraps it in a terminal experience that orchestrates whole pipelines instead of one-off prompts — for a small flat subscription that costs a fraction of what API-billed dev tools charge for the same work.

Install it, log in once, and you've got a local orchestrator that can plan, build, review, QA, and ship real GitHub PRs — plus a handful of focused tools for the parts of your day that aren't "build a feature."

Supported providers + verified models

Pick whatever fits your stack and your wallet. Each row is a path Claudio's worker pool will actually run end-to-end:

| Provider | How you sign in | Verified models that route through it | |---|---|---| | Claude Code | Existing claude auth (no API key) | Sonnet 4.6, Opus 4.7 | | ChatGPT Subscription | OAuth — sign in with your ChatGPT Plus / Pro / Team account | GPT-5.5, GPT-5.5 Pro, GPT-5.5 Codex | | OpenRouter | One API key for everything | Sonnet 4.6, Opus 4.7, GPT-5.5, DeepSeek V4 Pro, MiMo v2.5 Pro | | Anthropic (direct) | Anthropic API key | Sonnet 4.6, Opus 4.7 | | OpenAI (direct) | OpenAI API key | GPT-5.5, GPT-5.5 Pro | | Amazon Bedrock | AWS credentials | Sonnet 4.6, Opus 4.7 | | GitHub Copilot | Copilot subscription | Sonnet 4.6, Opus 4.7, GPT-5.5 | | DeepSeek | DeepSeek API key | DeepSeek V4 Pro | | Fireworks | Fireworks API key | DeepSeek V4 Pro | | Cloudflare AI Gateway | Cloudflare key | Sonnet 4.6, Opus 4.7, GPT-5.5 | | Vercel AI Gateway | Vercel key | Sonnet 4.6, Opus 4.7, GPT-5.5, DeepSeek V4 Pro | | Hugging Face Inference | HF token | DeepSeek V4 Pro | | OpenCode Zen / Go | OpenCode key | Sonnet 4.6, Opus 4.7, GPT-5.5, DeepSeek V4 Pro, MiMo v2.5 Pro | | Xiaomi MiMo | Token plan | MiMo v2.5 Pro | | Azure OpenAI Responses | Azure deployment + key | GPT-5.5, GPT-5.5 Pro |

Five verified families across ~15 distinct model IDs and ~30+ delivery routes. Provider list is a live filter — only providers with at least one verified model that we've tested end-to-end show up in the picker.

Install

pnpm add -g claudio-cli@latest

or

npm install -g claudio-cli@latest

or

yarn global add claudio-cli@latest

Then:

claudio-cli

The CLI walks you through gh auth + your chosen provider on first run. Pick Claude Code or ChatGPT subscription if you'd rather not paste an API key.

What it does

Feature pipelines

Describe a task. Claudio runs PM → Designer → Dev → Review → QA → Merge as a real, observable pipeline against a real GitHub repo. Every stage is a specialized agent; every stage emits structured progress that streams into a live TUI. The output is a merged PR, not a chat transcript.

Two ways to run them:

• Work mode — pick a batch of tasks, Claudio runs them in parallel (1–5 concurrent), re-plans as slots free up, and auto-retries stuck tasks up to 3 times before flagging them for you.
• One Shot — single task, single focused archer firing a constellation of commits. Fastest path from "I want this" to "PR is up."

Quiet Mode

Want the agents' horsepower without the GitHub noise? Flip any task to quiet mode and Claudio plans, builds, and reviews on a local branch — no issue, no PR, no push. The finished work lands on your Launch Pad where you decide whether to open a PR, squash-merge locally, or toss it. Great for spikes, sensitive code, or work you want to eyeball before the rest of the team sees it.

Launch Pad

A staging area that collects every piece of in-flight work — quiet-mode branches waiting to ship, PRs with merge conflicts to resolve, reviews ready to submit. Push them out in bulk when you're ready, drop into an interactive conflict resolver, or walk away and come back to them tomorrow. Everything an in-flight claudio session built stays here until you decide its fate.

Task Manager

Browse every task you've ever queued — filter by status (new / queued / running / blocked / retry / rejected), create new ones, edit long briefs inline, toggle between ship-a-PR and quiet-local modes, mark tasks as features or bug fixes, and block/unblock them without leaving the terminal. It's also where you recover stuck runs: anything the pipeline left half-claimed gets reconciled back to new on entry.

Doctor

Your UI feels AI-generated? The Doctor scans your frontend against the industry's sharpest design critique (visual hierarchy, typography, palette intent, layout rhythm, AI-slop anti-patterns), picks a visual direction that fits your project's declared voice, and hands the whole brief to a creative frontend-design agent with authority to rewrite pages, swap the token layer, and create the components a real revamp needs. Not a polish pass — a full aesthetic pivot, shipped as one PR you can revert as a unit.

Fresh projects

No repo yet? Claudio can spin one up end-to-end: it asks what you want to build, picks a stack, scaffolds the codebase on your machine, initializes git, creates the GitHub repo, wires in the agent scaffolding (CLAUDE.md, rules/, agent defs), and drops you straight into the pipeline so the first feature can start building while you're still reading the README.

Bug Hunt

Hand Claudio a bug report. It instruments the suspect code, reproduces the scenario, iterates up to 5 times adding more logs each round if the signal isn't enough, pins down the root cause, and lands a fix. Then it strips all the instrumentation it added so the PR is just the fix — no debug noise.

PR Review

Point it at any PR URL. Claudio drafts inline comments in a natural reviewer voice (no robotic "This function should…" fingerprints), cross-references comments other reviewers already left so it doesn't duplicate feedback, and hands the list back to you in a per-comment flow:

• Post — drop the comment inline on GitHub.
• View file — open the commented line directly on GitHub.
• Edit — tweak the draft before posting.
• Quick Fix — deploy a focused fix agent that commits AND pushes the change directly to the PR's head branch. Works from any directory, even if you've never cloned the PR's repo.
• Skip — not every draft is worth sending.

Designer

Describe a screen, a flow, or a whole section of your product and Claudio produces a real design for it — using your project's own colors, typography, spacing, and brand voice, not generic sample-app aesthetics. Works as single-shot or autopilot mode that keeps designing the next item on your list until it's done.

Memory Bank

Build a distilled project brief that any future agent can read as context. Captures the design system, the service layout, the code conventions, the "why" behind decisions — in a format the orchestrator injects into later prompts so your agents stop re-discovering the codebase on every run.

Imagination

Compares your product vision (ROAD.md) against the reality of your repo and generates net-new tasks that would close the gap. Optionally scope the dream to a specific slice of the codebase (packages/frontend, src/billing, "my backend only") — Claudio only proposes tasks inside that slice. Two modes — "just dream up tasks" or "dream then execute" — and a choice of horizon (day / week / month). Great for mornings where you know there's something to do but can't name it.

Why plug-and-play matters

Most AI dev tools bill per token. Every prompt, every retry, every "let me just check one more file" shows up on an invoice at retail API rates — plus the tool's markup on top. A full day of pipeline work can quietly run into double-digit dollars. A week of steady use, into real money.

Claudio is a flat, small subscription. No per-token meter, no surprise monthly bill, no quota anxiety mid-task. Whichever provider you pick — Claude Code, ChatGPT subscription, OpenRouter, an API key — Claudio just adds the pipelines and tools around it. If your provider is subscription-based (Claude Code, ChatGPT), you're not paying twice. If you brought an API key, you control the spend at the source.

The number you pay us is a fraction of what API-billed tools charge for equivalent work — by design.

Requirements

• Node.js 18+ (only used by the install-time shim that resolves the native binary)
• A worker provider — pick one:
  • Claude Code — npm i -g @anthropic-ai/claude-code + Claude.ai subscription (Pro / Max / Team), OR
  • ChatGPT Subscription — Plus / Pro / Team account, signed in via OAuth on first launch, OR
  • OpenRouter / OpenAI / Anthropic / DeepSeek / etc. — paste an API key once when the CLI prompts
• GitHub CLI (gh) installed and authenticated — gh auth login

Supported platforms

• macOS Apple Silicon (arm64)
• macOS Intel (x64)
• Linux x64
• Windows x64

Each platform ships as a Bun-compiled native binary — no Node source on disk, no sudo, no codesigning dance. npm install picks the right one automatically.

Links

• Website — https://claudiopipe.com
• Issues / support — [email protected]

License

MIT