I build AI applications that act, verify, and recover—
with a focus on local-first systems, agents, human control, and reliable execution.
AI Agents · Local-first · MCP · RAG · Human-in-the-loop · AI Security
- NEXORA / 知序 — An open-source, local-first personal intelligence platform with confirmed memory, a private Markdown Vault, and permission-controlled MCP access.
- Schedule Guardian — A human-in-the-loop agent for conflict-aware, verified schedule changes.
- Accessible Beauty AI — Real-time, accessible AI guidance for makeup practice.
- LvLing Travel Agent — An AI travel agent and digital-human platform.
- Tencent/teamai-cli — Added read-only session adapters for Pi / Oh My Pi (#1586), Muse Code (#1680), Hermes (#1705), and GitHub Copilot CLI (#1785); also preserved delivery records across equivalent HOME paths (#1780) and improved archive publication reliability (#1327, #1375, #1382).
- zalando-incubator/agentic-identity-broker — Added a dedicated, backward-compatible client-assertion audience for token exchange (#283).
- Tencent/WeKnora — Fixed CSV/TSV preview encoding detection by rejecting invalid UTF-8 before falling back to GBK (#4479).
- morluto/rea — Preserved the effective browser-bundle origin scope in generated evidence (#4413) and isolated real CLI subprocess deadlines for clearer, more reliable Windows test diagnostics (#5019).
- bytedance/deer-flow — Preserved timezone offsets correctly when persisting run-event timestamps in SQLite (#6894).
- Future AGI — Preserved picker-provided string provenance and routed JSON-/boolean-looking strings through a bounded Users matching walk (#3786).
- GitHub Advisory Database — Contributed evidence-backed advisory corrections and reference updates (#10998, #10999, #11000).
- OpenSSF Scorecard — Updated GitLab merge-request analysis to use the latest returned commit date without reordering results (#5339).
Build useful AI. Keep humans in control. Verify the result.