We build the memory agents need.
Arc Labs builds composable, open-source infrastructure for agent cognition — memory, planning, reasoning. Not wrappers around general-purpose databases. Not retrieval bolted onto a vector store. Brain, our first product, is a typed memory layer for AI agents.
Open source · Apache 2.0
Agents, not chatbots.
The next generation of AI applications will not be chatbots. They will be agents — systems that remember across sessions, plan multi-step workflows, reason about constraints, and take action through tools.
Today, every team building agents reinvents the same infrastructure: a vector store for retrieval, a database for state, a pipeline for filtering noise, a graph for entities. The result is fragile, untested glue code that breaks in production. We are building the standard instead — starting with memory, because an agent that can’t remember can’t plan or reason.
Read our story →One product today. Room for more.
Brain is our first product — the memory layer, because an agent that can’t remember can’t plan or reason. More agent infrastructure is on the way.
Brain
A typed memory layer for AI agents — Rust core, Apache 2.0, self-hosted or cloud. Scores 85%+ on LongMemEval with a write pipeline that rejects noise before it’s ever stored.
More agent infrastructure in development.
What we commit to.
These aren't marketing lines — they're constraints on every decision we make, from the API surface to the pricing page.
Open core, Apache 2.0.
The engine is free, forever. We make money on hosting, not by paywalling basic features. Nothing that ships free today will ever become a paid tier.
Provenance is non-negotiable.
Every memory links back to its source turn. Every retrieval is auditable. Every reasoning step cites its evidence. No black boxes.
Transparent pricing.
No 'basic feature is $249/mo' trap. LLM costs pass through at zero markup. If you can read a spreadsheet, you can predict your bill.
What we’re learning in public.
Why 98% of agent memories are junk — and what to do about it.
A 10,000-entry audit of production memory stores. What gets stored, what shouldn't, and the pre-write filter design that changes the ratio.
Typed memory vs flat text: a benchmark on LongMemEval_s.
Brain's typed schema vs Mem0's flat extraction on the same 500-turn eval. Where structure wins, where it doesn't, and why we still chose it.
Three conversations with indie devs building agent memory.
What they hate about Mem0. What they rolled themselves. Why most of them said 'I'd pay for this if it was self-hostable.'
We’re hiring. Systems and infrastructure engineers, Bangalore.
Open roles →