abird-ai/agentc is an embedded coding agent written in freestanding C23 that does not depend on libc. The repository description introduces it as a statically linkable, embeddable AI agent under 1MB. The latest release is v0.8.0, but the specific changes in this version are not contained in the provided record. So this entry is limited to the implementation and verification details the README reveals.
The README table lists roughly 940KiB for the stripped binary, 0.3ms for cold start, and 0.7MB for cold-start RSS. The same table also carries comparison figures for codex 0.161.0, Claude Code 2.1.293, and pi 0.99.2. The author describes an architecture with no interpreter and no garbage collector, and writes that it uses roughly 0.4MB of its own text and 0.2MB of heap while waiting on prompts. It prioritizes MCP and offers a versioned C ABI, and the README also specifies cross-builds and test entry points for six targets across Linux, macOS, and Windows.
That said, the footprint and startup figures are the author's own measurements, and the measurement method plus the versions and machine used are limited to the README. No independent reproduction report is contained in the provided record. So the small resource usage is an interesting implementation direction, but whether other environments produce the same results needs to be checked separately.