Persistent context, searchable history, entity tracking, and decision recall — across every session. One API. Two function calls. Ship agents that remember.
from codeclaw import Memory mem = Memory(api_key="cc_live_...") # After each agent turn → store what matters mem.store( user="u_0x7a3", content="Budget approved at $40k. Chose Vendor B. Prefers async updates.", entities=["project_atlas", "vendor_b"] ) # Before each turn → recall what's relevant ctx = mem.recall(query="What vendor did they choose?", user="u_0x7a3") # → "Chose Vendor B." (4 sessions ago · confidence: 0.97)
Works with your stack
You've built the reasoning, the tool calls, the chains. But your agent still forgets who it's talking to after every session. You patch it with JSON dumps, oversized system prompts, and raw vector queries. It doesn't scale. It breaks in production.
The user explained their project scope on Monday. By Wednesday the agent asks again. Users lose patience. They stop using it — not because the reasoning is bad, but because the memory is non-existent.
It recommended Plan A last week, Plan B today. What changed? What context did it use? There's no audit trail. You can't debug a decision you can't inspect. Production agents need accountability.
You've tried: vector DB + embedding pipeline + retrieval logic + per-user scoping + TTL policies. Three weeks in and it still surfaces stale context. This is infrastructure work — and it's not your product.
CodeClaw sits between your agent and durable recall. It handles storage, retrieval, scoping, entity linking, and temporal ranking. You call two functions. Your agent remembers everything.
Memory survives session boundaries, restarts, and deploys. Your agent picks up where the user left off — days or months later. No context window tricks.
Ranked by meaning, recency, and confidence — not keyword match. The right context at the right time. Nothing stale. Nothing irrelevant.
Memories link to people, projects, accounts, and decisions automatically. Query by entity: "What does the agent know about Project Atlas?" Get a structured answer.
Every decision logged with source context and timestamp. When someone asks "why did it say that?" — you have the answer. Inspectable in the dashboard or via API.
Memory isolated per user, workspace, and tenant. Architecturally enforced — not query-filtered. Zero cross-contamination from the first API call.
REST API. Python and JS SDKs. Two calls: store() and recall(). p95 under 50ms. Works with any model, any framework, any agent architecture.
No infrastructure to provision. No embedding pipeline to build. Install the SDK, add store() and recall(), deploy. Your agent has memory.
pip install codeclaw
One package. Zero config.
After each turn, call mem.store() with user, content, entities.
Before each turn, call mem.recall() to inject relevant history.
Use the dashboard to view, search, and debug every stored memory.
Any LLM · Any framework
store() · recall() · search()
Indexed · Scoped · Persistent
Your agent calls the API. CodeClaw handles indexing, retrieval, scoping, and ranking. You inspect via dashboard.
Recall from days, weeks, or months ago
Find any memory by meaning
People, projects, decisions — linked
Full audit trail for every action
Scoped per user by default
A lightweight dashboard for visibility and debugging. Search memories, inspect entity graphs, view decision timelines, and trace what context was used in any recall. Built for developers who need to understand what their agent knows.
Filter by user, entity, time range, or content. Find exactly what the agent stored and when. Semantic and keyword search.
See every memory linked to a specific person, project, or account. Understand what your agent knows about each entity across sessions.
Trace how agent decisions evolved over time. View what context was retrieved, what was stored, and what changed between sessions.
If your agent talks to the same user more than once, it needs memory. These are real scenarios where teams integrate CodeClaw today.
A user says "I already explained this." Your agent pulls up the prior thread, the resolution, and the pending follow-up. No re-explaining. Resolution time drops. CSAT goes up.
Your ops agent remembers that deploys to staging fail on Fridays. Your procurement agent recalls the client prefers net-60 terms. Agents that have seen the pattern act on it.
Remembers your OKRs, your last investor call, that you moved launch to Q2, that you hired for the design role. Every conversation builds on the last because nothing gets lost.
We're onboarding the first cohort of builders. Leave your email — you'll get early API access, free usage during beta, and a direct line to the team.
Be the first to give your agents memory.
No spam. Early access + product updates only.
First in line when we open the API. Start building before everyone else.
Full access at no cost while we're in beta. No credit card needed.
Shape the product. Early builders get a direct channel to the team.
recall() is fast even with hundreds of thousands of stored memories. No cold start. No batch processing. Context is available the moment your agent needs it.pip install codeclaw, add mem.store() after each turn, add mem.recall() before each turn. No embedding pipeline. No retrieval tuning. No infrastructure to manage.We're launching soon. Get early access and start building before everyone else.