The story behind
Recall Racoon
A deterministic cognitive control plane for LLM systems. Built by engineers who hit the same walls you're hitting.
Once upon a
raccoon...
The Raccoon's Problem
There was once a raccoon who could remember that he had carefully hidden his marbles. The problem was remembering where he put them. Every day became filled with raccoon responsibilities—trash to rummage through, interesting objects to investigate, nighttime adventures, and the perpetual need to keep his famous dark eye mask looking just right. Through all the hustle and bustle, he would remember that the marbles had been stored safely somewhere. He simply could not recall where.
The Lesson
Saving something is not the same as being able to recall it when it matters. The raccoon had a Stash (the hiding spot) but lacked Rumi (the intelligence to organize and retrieve). He had Rummage (the searching) but no Recall (the finding).
The Idea
That became the idea behind Recall Racoon. Businesses and AI systems create information constantly. The problem is not simply storing it. The problem is understanding it, organizing it, connecting it, governing it, and making it useful later. Rumi handles the hustle. Rummage processes the incoming stream. The Stash remembers. Recall retrieves.
Rumi works in the background so your people and AI systems do not have to become memory librarians.
Rumi does not replace your AI tools. Rumi helps them remember.
Rumi. Rummage.
Stash. Recall.
Rumi
The intelligence layer
Examines incoming information, identifies what matters, prepares it for the Stash.
Rummage
What Rumi does
Understand, summarize, categorize, connect—incoming info becomes organized memory.
The Stash
The memory layer
Persistent, searchable, governed. Claims with lifecycle, citations, provenance.
Recall
Retrieval when needed
Hybrid search, graph context, freshness scoring. Sub-50ms latency.
How we build
and why
Deterministic over probabilistic
Rules, not guesses. Same input = same decision. Auditability is architecture.
Local-first, cloud-optional
Run fully offline. Add cloud when you need scale, not because you're forced to.
Governance at write time
Validate at ingest. Bad memory never enters the Stash. Probes: format, citation, semantic, contradiction.
Your data, your rules
We never see your Stash. Architecture enforces this, not policy. No training on your data.
MCP-native, not wrapper
Built on the protocol, not around it. 13 tools. stdio, SSE, HTTP. Zero config.
Capacity-based, predictable
No per-token surprises. No unlimited promises. Transparent limits and overage rates.