The only pure-C LLM coding agent — and why that matters
Cursor, Claude Code, Aider, Antigravity — every autonomous coding agent today ships on Python or Node. That's fine on a developer laptop. It's a dead end everywhere else: the embedded box, the air-gapped CI runner, the minimal container, the bare-metal EPYC with no venv you trust.
Libcage is the exception. A 21KB binary, zero dependencies, libc + POSIX sockets only. It calls any OpenAI-compatible endpoint, reads your broken file, applies the fix, compiles, and tests — in a loop — until it's green.
libcage "Fix the off-by-one in parse()" broken.c \ "cc -o out broken.c" "./out" # iteration 1/5 # compiling... # testing... # SUCCESS after 1 iteration(s)
Why pure C is the right substrate
An agent that repairs code is only useful where the code actually lives. And code lives in places that can't run a Python interpreter with 400 transitive packages.
- No supply chain. One file, auditable by eye. No npm, no pip, no CVE lottery.
- Portable to the edge.
cc -O2 -std=c11on Linux, macOS, BSD, and minimal containers. - Fast to deploy.
curl | sha 21KB binary onto a box and it's running — no install step, no side effects. - Future-proof. C outlives every language trend. This binary will run in 2035.
Google's Antigravity is the fastest way to write agents. Libcage is what you ship when the agent has to run where Antigravity can't.
Built with Antigravity, shipped as C
Libcage was scaffolded and iterated using an LLM code-edit IDE (the "vibe coding" loop). The runtime itself, though, is dependency-free C — so you get the leverage of LLM-assisted building without the runtime baggage. The agent writes the C; the C runs the agent. That's the dogfood loop, proven end-to-end.
Two tiers
Libcage ($29, one-time) — the autonomous repair agent. Get Libcage →
Libcage Pro ($99, one-time) — adds CycloneDX SBOM generation and team endpoint-policy enforcement for regulated shops. Get Libcage Pro →
What's next
A Team tier (multi-seat license + shared policy) is on the way, plus automated GitHub releases so each tag builds and publishes the static binary. Watch the repo: github.com/Hardonian/libcage.