Your AI Coding Assistant
Drive AI coding assistants via browser automation. No API costs. Works with ChatGPT, Claude, Gemini, and Browser.
Get Started in Seconds
One command to install FreeCode and start using AI coding assistants.
$ curl -fsSL https://freecode.website/install | bashFreeCode is a harness that is trying to push the frontier of intelligence. We believe that the next leaps in agent capabilities will come from the harness level. The model is just the intuitive core of an agent, and the harness around it exists to extract as much useful work out of that core as possible. Scaling test time compute via agent swarms, increasing speed through parallelism, solving continual learning through memory. We publish everything publicly, and will remain open source and open research. At the same time, we strive to be a good general purpose dev tool for day to day work.
Swarms only push intelligence if they scale, and they only scale if each agent costs almost nothing. FreeCode is optimized to the bone so agents stay cheap to spawn, with none of it traded for speed.
Memory footprint (PSS, MB)
Resident memory of the entire process tree (tool + all descendants + process group), summed across process descendants.
Time to ready (seconds)
Startup latency: wall-clock from spawn until input handling is responsive (10 interactive PTY launches each).
codex, opencode and copilot_cli hit the 20s timeout without becoming input ready; antigravity_cli never rendered visible.
Memory per session, avg over 10 runs (MB)
Extra proportional memory (PSS) each additional client adds once one is already running. Ten freecode sessions cost about 250 MB, less than one sixth of one Claude Code.
Tokens are the bill. On a large tool output whose answer is in the tail — a build error at the end, the last grep matches — FreeCode injects the fewest tokens, capturing the full output once and letting the model page the rest instead of re-running the command.
Model-input tokens per large-output turn
Lower is better. Tokens each tool’s harness sends the model to get head + tail of an oversized output. Modeled from each tool’s truncation policy in its own source — not a live API run.
How to read this: FreeCode’s lead is a tighter first-turn cap (30 KB vs opencode/pi’s 50 KB) plus an in-memory output store — frugal by design, not strictly more capable. opencode and pi also have retrieval (file-based), so they avoid the re-run penalty that the head-only tools pay. Rows marked * are unverified from source. Modeled, deterministic, no API calls.