OPEN LINK ↗
// 1h agoOPENSOURCE RELEASE
Caveman 3.0.0 ships Apache-2.0, Learn autopilot
Caveman 3.0.0 turns its viral token-saving skill into broader developer infrastructure, relicensing the repository under Apache-2.0 while adding autonomous Learn audits, stable middleware, SDK 1.2, and production proxy features. The release also introduces an SDK license change and default telemetry, making migration review essential. [Release notes](https://github.com/JuliusBrussee/caveman/releases/tag/v3.0.0)
// ANALYSIS
Caveman is evolving from a clever prompt-compression stunt into a serious context-efficiency layer for coding agents, with unusually strong attention to measurement and reversibility.
- –Apache-2.0 across the engine, proxy, MCP server, shrink tools, and memory core removes hosted-service restrictions and makes enterprise embedding easier.
- –Learn’s autopilot, memory doctor, trend tracking, and rollback-if-no-savings workflow turn token optimization into an ongoing maintenance process.
- –Middleware 1.0, SDK 1.2, fail-open adapters, Postgres HA, TLS/mTLS, and deployment manifests broaden Caveman beyond terminal-based agent wrappers.
- –Telemetry is enabled by default for the CLI and hooks, sending usage statistics, a random install ID, and IP address; teams should disable it where policy requires.
- –The project’s claimed savings still need workload-specific validation: its Product Hunt listing cites a pinned 54-run benchmark with 33.2% fewer provider-reported input tokens and 18/18 correctness checks. [Product Hunt listing](https://www.producthunt.com/products/caveman)
// TAGS
cavemanai-codingcoding-agentagent-memorymcpcontext-engineeringopen-sourceself-hosted
DISCOVERED
1h ago
2026-10-02
PUBLISHED
1d ago
2026-09-30
RELEVANCE
9/ 10