The archive · Developer & Business Tools · Product decision · 2024–2026
Memori bets agent memory should record actions, not just conversation
New York startup's Apache-2.0 memory layer for AI agents hit 11.1K GitHub stars by 2025-12-18 on $3.7M raised.
Memori Labs
What the business is
Memori Labs (New York, founded 2024) sells an agent-native memory layer: Python and TypeScript SDKs plus an MCP server that record what LLM agents do — execution traces, tool calls, decisions, outcomes — into structured persistent memory and recall it on demand. The core is Apache-2.0; revenue comes from Memori Cloud and private-VPC Memori Enterprise deployments.
How it started
Memori Labs was founded in New York in 2024 on the observation that agents suffer session amnesia: every new session forgets the stack, the conventions and what earlier runs already established, so teams repeat themselves and agents re-discover what prior work found. The founders bet the fix was a memory layer that records what agents do — tool calls, decisions, outcomes — not just what they say.
What happened
The team shipped an Apache-2.0 core with Python and TypeScript SDKs plus an MCP server, so Claude Code, Cursor, Codex, Warp, OpenClaw and Hermes agents get persistent memory without prompt changes. Memori Cloud offers zero-configuration hosted memory with bring-your-own-database options, and Memori Enterprise runs inside a customer's VPC; the README cites one enterprise QA-automation deployment cutting token spend by $2.1M per year. ROSS Index Q4 2025 recorded 11.1x GitHub star growth over the quarter, on top of a $3.7M round from F7, Oceans, RiverPark and Struck.
No ending yet — it is still running.
Background
Memori Labs (New York, founded 2024) builds an agent-native memory layer for LLM applications. Its Apache-2.0 core captures execution and conversation into structured persistent state — tool calls, decisions and outcomes — and makes it recallable by agents through SDKs or an MCP server.
The founding wager is that agents fail from amnesia, not capability: they forget the stack, the conventions and what an earlier run established, so teams repeat themselves and burn tokens re-discovering what is already known. Memori's answer is memory built from actions rather than chat text, stored as structured facts, preferences and rules instead of raw logs.
By Q4 2025 the bet had traction: ROSS Index counted 11.1K GitHub stars on 2025-12-18, an 11.1x increase in the quarter, with $3.7M raised from F7, Oceans, RiverPark and Struck. The project's LoCoMo benchmark reports 87% overall accuracy at an average 721 tokens per query — roughly 2.8% of the full-context footprint.
The open core feeds a paid stack: Memori Cloud for zero-configuration hosted memory and Memori Enterprise for private-VPC deployments, with integrations for Claude Code, Cursor, Codex, Warp, OpenClaw and Hermes. The README cites one enterprise QA-automation workflow reducing token spend by $2.1M per year.
What has to be true
- Session amnesia is the common failure: agents forget conventions and prior results between sessions, so the same discovery work repeats and token spend compounds.
- Recording what agents do — tool calls, decisions, outcomes — captures context that conversation text alone misses, which is the differentiator behind the agent-native positioning.
- Structured recall beats full-context replay: 721 tokens per query at 87% LoCoMo accuracy makes persistence cheaper than stuffing history into every prompt.
- Open-sourcing the core under Apache-2.0 while selling cloud and enterprise deployment removed adoption friction, driving 11.1x GitHub star growth in one quarter.
What can be applied
Memory wins by capturing what agents do, not just what they say: tool calls and outcomes are the context they need, and recalling a slice beats replaying everything.
Aftermath
As of September 5, 2026, Memori is live with a hosted cloud, bring-your-own-database support and an enterprise tier, integrated as an MCP server for Claude Code, Cursor, Codex and Warp and as plugins for OpenClaw and Hermes. Its README reports 87% accuracy on LoCoMo at 721 tokens per query and cites a $2.1M-per-year token-spend reduction from one enterprise QA-automation deployment. With 11.1K GitHub stars on 2025-12-18 and $3.7M raised from F7, Oceans, RiverPark and Struck, Memori Labs is scaling the open-source community into paid enterprise memory contracts.
Sources
- Top trending open-source startups in Q4 2025 — ROSS Index
- Memori README — Memory from what agents do, not just what they say
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card