Apache-2.0 · Local-first · Built in Rust
A local-first agentic coworker you can configure without limit. Point it at your files, choose a model, and get a working spreadsheet, a deck, or a triage app you can reopen.
Windows + Linux coming soon
Sign in with ChatGPT. The OAuth handshake runs on your own machine, and you use the subscription you already pay for. No API key to create.
Anthropic, OpenAI, Google, xAI, or OpenRouter. The key goes in the macOS keychain, not in a config file you might commit.
Point the OpenAI-compatible slot at Ollama and run open-weight models on your own hardware, so even prompts stay on the machine.
Start on the ChatGPT plan you already pay for.
Anthropic, OpenAI, Gemini, xAI, OpenRouter, or a local Ollama model. Switch mid-chat. Fail over. Cheap model for the grind, strong model for the hard step.
Same exec tool. Different backends.
Per-chat network policy: off, package managers, allowlist, or open. Credentials never enter the sandbox.
A folder is available only after you pick it.
Overwrite of a connected file always asks, in every permission mode.
Computer use (screen and app control) is partial.
Look things up without leaving the machine.
Files plus code execution are the interface to your data. MCP and REST come later.
Finished files you can preview, version, and hand to someone.
A spreadsheet, a deck, a memo. Not a wall of markdown.
The engine runs on your machine. The defaults stay boring on purpose.
Rust, because the loop is fast enough to feel like a native tool.
Same filename appends a version. Restore appends too. The file stays on your disk, in the Outputs catalog, until you Save As.
A native picker, attached to that chat. The model never sees an absolute path. A new chat inherits nothing.
The journal is provider-neutral. Switch mid-task when a provider rate-limits you, or when the next step needs the expensive model. The conversation does not restart.
Plan is read-only. Ask parks every mutating call. Auto is a standing yes to workspace edits. Allow all is an opt-in. A file you wrote yourself always asks, in every mode.
For three years we built enterprise-grade research agents for some of the most high-fidelity work there is: private equity diligence, the kind of file someone stakes their career on.
Now we're shifting our focus to secure, compliant infrastructure for the agentic AI era. Tidebreak is the open coworker that comes out of that.
Interfaces and local data still change between releases. See what is live today, what is next, and what we are still exploring.
Read the roadmapmacOS Universal. Your ChatGPT plan, a key, or Ollama. No Tidebreak account.