The research behind the product
We share the research and tooling we build to solve hard problems in AI. These projects power what we do at Parslee — most are fully open source; some ship as freely-available binaries with public docs.
View on GitHubStateBench
The benchmark and engine for AI memory that actually works
AI systems claim to have memory. They do not. They replay transcripts and hope for the best. StateBench exposes the failures — resurrection, stale reasoning, scope leaks — across thirteen tracks. Memgine, our deterministic memory engine, then shows how to fix them: by enforcing correct state at the architecture level, not the prompt level. Two published papers, one open source repo.
- Thirteen failure tracks targeting distinct memory failure modes
- Memgine engine substantially outperforms transcript-replay baselines
- Two published papers: architecture (2025) and engine (2026)
Neo
The memory layer for AI coding tools
Every developer knows the frustration: your AI assistant forgets what you just told it, repeats the same mistakes, ignores your project's patterns. Neo fixes this. It's a reasoning engine that actually learns—building a knowledge base from every coding session that makes future suggestions smarter. Think of it as giving your AI assistant a memory that persists.
- Learns from both successes and failures
- Works with OpenAI, Anthropic, Google, or local models
- Capable of running locally with on-device models, or routing to OpenAI, Anthropic, Google, or Azure — you choose where your code goes
MCP-API
Turn any REST API into an AI tool
AI agents need to talk to APIs, but every API speaks a different language. MCP-API is the universal translator. Point it at any OpenAPI spec, Swagger doc, GraphQL schema, or Postman collection, and it instantly becomes a tool your AI can use. No custom integrations, no manual configuration—just connect and go.
- Supports OpenAPI 3.x, Swagger 2.0, GraphQL, and Postman
- Built-in auth handling (API keys, OAuth2, bearer tokens)
- Pluggable secret storage; supports Azure Key Vault, environment variables, or file-based config
CAR (Common Agent Runtime)
A deterministic execution layer for AI agents
Agents that pass raw LLM output straight into tool calls fail unpredictably — unsafe actions, hallucinated tool names, state drift. CAR treats plans as first-class data: verify before executing, enforce policies in Rust before any side effect, track memory as a graph, and learn reusable skills from execution traces. Neo routes its outbound inference through CAR's adaptive router.
- Static verification of plans (verify / simulate / equivalent / optimize) before any side effect runs
- Rust-enforced policy engine on every action; graph memory with spreading activation
- Distributed as public binaries with full docs; runtime source is currently private
Why we share our work
Building reliable AI is hard. By publishing our research and tooling — most fully open source, some as freely-distributed binaries — we help the whole industry move forward and show that we understand these problems well enough to solve them for our customers.