epistemic_store

Store a natural-language claim as a validated, scored, and classified memory.

POST

epistemic_store

Description

Accepts a raw text claim and processes it through the full 6-layer pipeline: policy check → sentence classification → normalization → confidence scoring → conflict detection → embedding & storage → tier routing.

Parameters

Name	Type	Required	Description
`claim`	`string`	Yes	The natural-language claim to store (e.g., "I work at Google").
`source`	`string`	No	Override source. One of: `user_explicit`, `agent_inferred`, `third_party`, `system`. Default: auto-detected.
`decayClass`	`string`	No	Override decay class: `PERMANENT`, `STABLE`, `MODERATE`, `EPHEMERAL`, `VOLATILE`. Default: auto-classified.

Example

Request

{
  "tool": "epistemic_store",
  "input": {
    "claim": "Tôi thích ăn phở bò Hà Nội"
  }
}

Response

{
  "success": true,
  "memory": {
    "id": "mem_f8a2b1c3",
    "claim": "Tôi thích ăn phở bò Hà Nội",
    "subject": "user",
    "predicate": "likes eating",
    "object": "Hanoi-style beef pho",
    "confidence": 0.741,
    "tier": "WORKING",
    "kind": "preference",
    "decayClass": "STABLE",
    "source": "user_explicit"
  },
  "pipeline": {
    "l0": "PASS",
    "l1": "normalized",
    "l2": "scored:0.741",
    "l3": "no_conflict",
    "l4": "embedded",
    "l5": "tier:WORKING"
  }
}

Errors

Error	Cause
`POLICY_BLOCKED`	Claim contains PII or violates policy rules
`NOT_STORABLE`	Sentence classifier determined content is not a factual claim
`ENTROPY_HALT`	System entropy too high — resolve conflicts first
`EMBEDDING_FAILED`	OpenAI API error during vector generation

Notes

The pipeline processes claims in <200ms typically
Vietnamese and English are both supported natively at L1
If a conflict is detected at L3, both the old and new claims are preserved with appropriate tier changes