There’s a quiet economic tension in agentic AI right now: you’re probably already paying ~$20–200/month for a Claude Code or Codex coding plan, and then every new “AI agent” app wants you to plug in separate metered API keys and pay again for the same model access. Cindy is the first open-source agent I’ve seen built specifically to collapse that double-charge.
The Core Idea #
Cindy is an open-source (Apache-2.0) AI agent that works out of the box. It runs locally on your own machine, on your real files and logged-in apps, and brings multiple harnesses, models, and tools into one agent. The headline:
Authorize the Claude Code / Codex coding plan you already pay for and keep using it inside Cindy — no duplicate bill.
So the same subscription that powers your terminal coding agent now powers a general-purpose assistant that can drive your browser, computer, and phone, and take work from messaging and schedules. You can also sign into Cindy’s hosted service, bring your own API keys, or run local models — your choice per task.
The Genuinely Interesting Architecture #
Beyond the billing angle, the design is what we keep arguing is the right shape for agents — a harness-agnostic layer that owns memory and skills while delegating execution:
- Harnesses and models mix freely and switch mid-task while your workspace, memory, skills, and tools stay continuous. Claude Code for one step, Codex for another, a local model for a third — same context throughout.
- One task can be planned, executed in parallel, and reviewed by agents on different harness × model combos (their “Orca” multi-agent orchestration). Plan with a strong reasoning model, fan out execution, review with a different model —the multi-tier org-chart pattern made concrete.
- Memory shared across harnesses — correct her once and it sticks everywhere.
- Skills, MCP, plugins, automation — teach a workflow once, wire in internal tools, schedule recurring work.
Tool binaries (claude-code, codex, ripgrep) aren’t committed — they’re downloaded per platform at install, sha256-verified. Clean supply-chain hygiene.
The Honest Caveat #
It’s open-source client, hosted backend. The README is upfront: the backend service lives in a separate, non-public repo, and the client connects to Cindy’s official cloud by default for server-backed features and auto-updates. “Skip Sign-In” runs local agents without an account, but you lose the server-backed capabilities. So this is not a fully self-hostable stack the way some would like — it’s a transparent, auditable, forkable client on top of a commercial service. For the subscription-reuse feature specifically, that’s a reasonable trade.
Why It Matters #
Cindy is a strong signal of where agents are heading: decoupled from any single model vendor, continuous across harnesses, and respectful of the subscriptions you already hold. The “don’t make me pay twice for the same model” pricing stance is the kind of thing that wins users, and the mix-and-switch-mid-task orchestration is exactly the flexibility that fixed single-harness agents lack.
If you live in Claude Code or Codex but want a general assistant that reuses that plan — and you’re comfortable with a hosted backend behind an open client — Cindy is well worth a look.
**Repo:** [github.com/makecindy/cindy](https://github.com/makecindy/cindy) · **Site:** [cindy.app](https://cindy.app)