Organisational memory
for AI agents
Part of the CueCrux platform
Unified, versioned, auditable memory that sits between your AI agents and the context they need to act safely. No more fragmented knowledge. No more agents without institutional judgment.
Every answer scoped to a token budget. Every state change checked and receipted.
The agent memory wall
AI agents are measured in hours. Human jobs span years. The institutional context that makes senior employees valuable has no machine-readable representation. Until now.
Fragmented knowledge
Conversations, context, and decisions are siloed across OpenAI, Claude, Slack, and a dozen other tools. No single source of truth.
Context that lives in heads
The knowledge that prevents agents from causing damage (which database is production, which vendor terms are unwritten) exists only in senior employees' heads.
Agents without judgment
75% of frontier models break previously working features during maintenance. Not because they lack capability, but because they lack organisational context.
This isn't a hunch. It's measured.
More context isn't the answer — at production scale it makes strong models worse. ScoreCrux, our open agent benchmark, shows why tool-mediated memory wins on both accuracy and cost.
Core recall at 2M-token scale
higher is betterFind 25 decisions + 5 needle facts across 3,346 documents — the Delta corpus, ResearchCrux.
More context makes strong models worse — and costs up to 6.6× more per run for far worse recall.
Context tokens for the same answer
lower is betterNear-identical score; a fraction of the token bill.
~6.7× fewer tokens, and most accurate at resolving which fact was current — ScoreCrux context benchmark, claude-sonnet-5 arm.
How it works
MemoryCrux wires existing VaultCrux platform capabilities into a unified, agent-friendly MCP interface.
Connect your platforms
Provide read-only API keys for OpenAI, Claude, and other sources. Keys are encrypted via Vault Transit and never stored in plaintext. Or skip keys entirely on the free local daemon: corecruxctl ingest ./docs.
Encode your judgment
Capture organisational constraints, decision context, and institutional knowledge. Natural language in, machine-checkable boundaries out.
Agents query via MCP
Every agent in your stack queries MemoryCrux before acting. One memory layer, 100+ tools, zero vendor lock-in.
Ready to give your agents a memory?
Install Crux locally and give your agents durable memory without creating an account.
Install Crux