Rust primitives for AI agents, LLM infrastructure, and financial data

I've been building the substrate layer for production AI systems in Rust. This is the index repo for 10 crates I shipped this week. The AI side covers agent memory (episodic, semantic, working with decay and multi-agent bus), cost governance with hard budget enforcement and automatic model downgrade cascade, CRDT state sync for distributed agent fleets, output versioning with full lineage tracking, a knowledge graph on top of agent memory, LLM inference primitives for WASM and edge runtimes, and a complete ReAct loop that runs inside a Cloudflare Worker. The financial side covers market primitives (orderbook, OHLCV, indicators, position, risk) and a streaming ingestion pipeline handling 100K+ ticks/second across 4 exchanges.

Everything is zero-panic enforced at the lint level, typed errors with thiserror throughout, no unnecessary dependencies. The crates are designed to compose, the orchestrator runs the agents, memory stores what they know, the graph understands relationships between what they know, the budget governs what they spend, sync keeps 24 agents consistent without locks.

The orchestration engine that coordinates all of this is tokio-prompt-orchestrator, which I've been running with 24 simultaneous Claude Code agents.

https://github.com/Mattbusel/rust-crates