squeezy documentation.
squeezy is a coding agent you run in your terminal. Install it, connect a model, then read the topic you came for.
contents
contents
Start here
New to squeezy? Three pages, in this order: install & upgrade to get it running, providers & models to connect the model you want it to use, and how it works to see what it does on your machine before a model is involved. Everything else is reference you can reach for when a question comes up.
squeezy also carries its own guides inside the binary, so the copy that
matches the version you installed is one command away. At a shell,
squeezy help lists the topics and
squeezy help <topic> prints one guide. In a session,
/help does the same, and /help /<command>
explains a single slash command. A topic or command squeezy recognizes
is answered from the docs on your machine: nothing is fetched and no
model is called. Ask a question in your own words instead and it goes to
your model, which costs tokens.
Help & troubleshooting covers that path,
and how to turn it off.
All topics
Install & upgrade
The install paths for macOS, Linux, and Windows, the first-start check with squeezy doctor, and how upgrades work.
conceptsHow it works
What squeezy does on your machine before it calls a model, and how modes, subagents, and model routing fit together.
costCost-saving strategies
The four layers that keep model spend down, and the ledger that itemizes what each one saved.
coverageLanguages
The 40+ languages and formats squeezy indexes on your machine, and what the code knowledge database can answer about them.
modelsProviders & models
Bring your own API key, or sign in with a ChatGPT or Copilot subscription, then pick the model each turn runs on.
toolsMCP & web
Add MCP servers to give squeezy more tools, and decide when it may search or fetch from the web.
verifyBrowser & computer use
Check a web app in a browser squeezy launches with its own profile, and drive a desktop app in a window you grant it.
extendSkills & prompt templates
Reusable instructions squeezy loads only when they apply, and your own slash commands, one Markdown file each.
lifecycleHooks
Run your own commands around prompts and tool calls. Anything a repository declares stays inert until you trust it.
settingsConfiguration
Where settings live, how the layers override one another, and how to keep API keys out of config files.
local dataStorage
Where squeezy keeps local data, how it tells one project from another, and what to back up or clean.
policyPermissions & safety
What squeezy may do without asking, how to change that, and what the shell sandbox blocks.
resumeSessions
How work is saved, resumed, and exported as a redacted bundle, and how to undo the agent's file changes once checkpoints are on.
integrateEditors, SDKs & CI
The same agent from an editor over the Agent Client Protocol, from your own code through the Rust, TypeScript, and Python SDKs, and CI state read through one interface.
supportHelp & troubleshooting
Ask squeezy about itself, diagnose a bad start with squeezy doctor, and send a redacted report when something fails.