AI-agent infrastructure, MCP developer tools, and secure Solana automation.
Projects · Current Focus · LinkedIn · X · Email
I build open-source infrastructure that makes AI agents more useful, observable, and safe: coding assistants, MCP discovery and memory, multi-agent operations, and transaction-capable Solana tools.
My strongest work is designed as software you can install and run—not just a concept demo. The flagship projects below have automated tests, documented security boundaries, and reproducible local setup.
| Project | What it does | Proof |
|---|---|---|
| deepseek-code | Low-cost, open-source AI coding assistant CLI with a real tool-use loop | Python package · 29 tests |
| clawdbot-deepseek | Self-hosted personal AI with persistent memory | Python · Docker · deployment guides |
| mcp-discovery | Offline MCP discovery over a bundled 17K-server index | TypeScript · 22 tests · hosted mode |
| sol-agent-wallet | Transaction-capable Solana agent with explicit spend controls | Python · 34 tests · security model |
| conductor | CLI, web cockpit, and MCP server for live coding-agent sessions | Zero dependencies · full test suite |
| gmem | Persistent project memory for Solana agents over MCP | TypeScript · SQLite · schema versions |
A growing set of small, offline, zero-dependency Python tools for checking what agents did — and what they're allowed to do. Each one reads a recorded trace, manifest or plan, never calls a model, and does one job.
| Area | Tools |
|---|---|
| Security & authority | agent-injection-scan · agent-memory-scan · agent-mcp-audit · agent-dep-audit · agent-permission-lint · agent-delegation-check · agent-cap-token · agent-egress-check · agent-pii-check |
| Approval & evidence | action-boundary · agent-approval-audit · agent-evidence-check · agent-receipt-verify · agent-effect-check · agent-citation-check |
| Reliability | agent-idempotency-guard · agent-operation-ledger · agent-retry-audit · agent-fallback-chain · agent-loop-watch · agent-budget-guard · agent-context-budget |
| Traces, plans & contracts | agent-span · agent-trace-check · agent-plan-lint · agent-tool-diff · agent-contract-check · agent-handoff-check · agent-eval-check · agent-novelty-check |
- Agent operations: observability and control for parallel coding agents
- MCP infrastructure: discovery, orchestration, and durable project memory
- Secure execution: bounded authority for agents that can move assets
- Agent reliability: offline auditors for traces, permissions, retries and evidence
- Open-source economics: practical alternatives to expensive AI tooling
- Local-first when practical
- Explicit security boundaries for write-capable agents
- Small dependency surfaces
- Automated tests for critical behavior
- Honest labeling of prototypes and experimental software
Python · TypeScript · JavaScript · Node.js · MCP · SQLite · Solana




