Cofounder/CEO
Profile summary: Stanford Biodesign Research Collaborator
Memorable
Memorable provides a procedural, graph-based memory system for AI agents, capturing successful runs and reusing them on future tasks to reduce latency, token usage, and error rates. The system stores deterministic, model-free procedures in the user's own local database and recalls them via task-description embeddings, supporting harnesses like Claude Code, Codex, Cursor, and OpenCode. It reports 19% fewer turns per task in GBrain on Claude Code.
- Developer Tools
- AI Agents
- Artificial Intelligence
- Software Only
Updated 9 days ago
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
across 1 round
$500K · Accelerator
Announced Jan 2026 · from article date
Disclosed cumulative: $500K
RoundDateAmountValuationInvestors
Founders
Product
Problem
AI agents typically start from zero on every task, even when they have successfully solved the same or similar problems before. This lack of persistent, structured memory forces agents to repeatedly perform costly exploration—reading files, searching repositories, and re-attempting failed approaches—which increases latency, token usage, and error rates while slowing down repeat workflows.
Solution
Memorable provides a procedural, graph-based memory system for AI agents that captures the steps from successful runs and reuses them on future attempts. When an agent finishes a session, Memorable's extraction layer converts the raw trace into a deterministic, model-free procedure that records the exact sequence of tool calls and conditions that reached the goal. These procedures are stored in the user's own local database and later recalled and handed back to the agent before it begins a similar task. The system assembles multiple procedures into a graph, where shared steps and prefixes reveal new, more efficient paths and allow the agent to compose workflows and develop implicit skills. Recall uses task-description embeddings to match the incoming prompt against stored procedures, raising the right workflow without requiring the agent to re-derive it from scratch.
Target Audience
Primary users are developers and engineering teams running coding agents such as Claude Code and Codex, as well as teams deploying browser, computer-use, voice, research, and operations agents that perform repeatable multi-step tasks and would benefit from persistent procedural memory.
Features
- Four-layer architecture: captures raw session traces, synthesizes successful runs into reusable workflows, assembles them into a graph connecting related procedures, and exposes a retrieval layer that agents query at runtime
- Deterministic, model-free extraction: the same trace produces the same steps in the same order, with no model in the loop able to rename, reorder, or edit the captured sequence
- Support for multiple agent harnesses including Claude Code, Codex, Cursor, OpenCode, and gbrain through session hooks and CLI backfill commands
- Recall matching via task-description embeddings, truncated to 2,000 characters, with optional local embedding for fully offline operation
- Graph assembly that connects workflows by shared steps and prefixes, enabling the discovery of new paths and composition of implicit skills
- CLI-based installation and management (`npx memorable-cli`, `memorable start`, `memorable enable`, `memorable recall`, `memorable show`) with hooks for automatic capture and recall on every new prompt
- Privacy-first design: coding-path procedures are written to the user's own store with no service-side copy kept, and source code and conversation transcripts are never transmitted
- Dashboard with live row counts, audit logs, retention controls, and evaluation runs for workspace administration