Memory · Procedural
Workflows you can run.
Procedural memory stores agent routines as executable units. Tag inputs, declare guarantees, and the runtime composes them — no orchestrator required.
Capabilities
What it does.
Every capability is production-grade. No flags, no betas — these ship the day you adopt Minds.
Executable
Every procedure is a callable function with typed inputs/outputs.
Versioned
Procedures carry a version, lineage, and rollback graph.
Composable
Chain procedures into pipelines; the runtime parallelizes safe steps.
Sandboxed
Each procedure runs with capability tokens scoped to its needs.
Consolidated
Recurring episode sequences are promoted to procedures automatically.
Observable
Per-procedure metrics: success rate, p50/p99 latency, retry rate.
API
A few lines is all it takes.
The SDK reflects the conceptual model directly. No glue code, no orchestrator, no learning curve beyond the data shape.
exampletypescript
// Define a procedure once. Run it forever.
await minds.memory.procedural.define({
name: "investigate-revenue-anomaly",
inputs: { quarter: "string" },
outputs: { cause: "string", confidence: "number" },
steps: [
{ tool: "sql", query: "SELECT * FROM revenue WHERE quarter = {{quarter}}" },
{ tool: "recall", modes: ["episodic", "graph"], query: "revenue anomalies in {{quarter}}" },
{ tool: "reason", prompt: "What caused the dip?" },
],
});
const result = await minds.memory.procedural.run("investigate-revenue-anomaly", { quarter: "Q3" });Specs
What it costs, what it guarantees.
Performance
- Latency overhead
- <2ms per procedure call
- Composition depth
- unlimited (cycle-detected)
- Concurrent procedures
- 10K+ per agent
- Rollback
- structural — every version preserved
Guarantees
- Durability
- fsync per WAL commit
- Isolation
- MVCC · snapshot
- Audit
- BLAKE3 Merkle chain
- Encryption
- AES-256-GCM per-namespace
- Concurrency
- 100K+ ops/sec
Build it on Minds.
Start with the SDK. Ship in an afternoon. Run anywhere — Akasha Cloud, on-prem, or air-gapped.