I build offline-first tools: local AI compaction, tiny decision engines, ESP32 firmware, and the docs to reproduce them. Everything runs on real hardware — Android/Termux, laptops, ESP32.
| Repo | What | Stack |
|---|---|---|
| local-compact | Free local compaction for Claude Code + OpenCode. No API key, honest tokens, receipt-guard, preview before you drop. | Python, 0 deps |
| tiny-decide | 5MB offline decision engine: typed choice/score/yes-no in <1ms CPU. Jev-compatible. | Python, 0 deps |
| nimbus-one | Autonomous AI agent in one static Go binary. No Node, no Python. | Go |
| raptor | Buffer-safety checks at the Rust↔C FFI boundary: canaries, poisoning, page guards. | Rust |
| bms-connection-tester | ESP32-S3 firmware impersonating a JBD Smart BMS over RS485. 32/32 tests. | C++, ESP-IDF |
| Repo | What |
|---|---|
| bms-tester-sim | Offline 3D bench simulator for the BMS tester build |
| ESP8266-Home-Automation | 4-channel relays + captive-portal WiFi + OTA |
| jbd-bms-rs485-handbook | RS485 + JBD protocol handbook with worked examples |
| esp32s3-qemu-arm64 | QEMU ESP32-S3 on ARM64 + flash-model fixes + Termux notes |
- CI passing (Python / Go / Rust / PlatformIO / Shell)
- License,
.gitignore, reproducible tests - No forks, no dead code
good first issue labels live on local-compact and tiny-decide — transcript adapters, receipt words, latency reports. PRs welcome.