ferrum-mcp
Workspace-scoped file, test, git, and optional crypto/DeFi/memory tools for AI agents over stdio or HTTP.
Side projects that don't live on the home page but are worth a look. Grouped by primary language. Featured work is on the home page.
Workspace-scoped file, test, git, and optional crypto/DeFi/memory tools for AI agents over stdio or HTTP.
Neo4j-inspired Rust property graph DB so agents can record, recall, and reuse the steps they take toward a goal.
HTTP clients offload fetch/compute jobs, get a job ID, poll for results. Tokio + axum, backpressure built in.
Per-API-key usage tracking, rate limits, exact-decimal invoices computed in parallel.
Unified FaError/FaResult, fixed-schema JSON logging, OTel span export with #[instrument_fa]. v1.0.0 on crates.io.
Off-chain execution via TransactionBatchProcessor, Solana L1 settlement, Celestia (mocked) DA.
Planning, execution, review, memory, and human approval in one FastAPI + LangGraph control plane.
TDD task loops with replay and dream consolidation for the Ferrum Platform.
FastAPI memory layer: working memory, hybrid retrieval, prioritized experience replay.
Correctness, safety, trajectory quality, BWT/FWT continual-learning metrics.
User profiles, versioned prompt assets, episodic interaction memory for agents.
Hybrid retrieval over personal technical libraries: Ollama + BM42 sparse + Qdrant RRF. Zero API cost.
Shared tools, schemas, guardrails, Docker sandbox, eval harness across LangGraph, OpenAI Agents SDK, CrewAI, ADK, Strands.
How RL, continual learning, world models, and plasticity repair are building the next generation of AI systems.