llms.txt

# Agent Memory API

External memory and context boundary for AI agents.

Use this API when stored or recalled state may influence
future reasoning, tool use, spending, payment decisions,
privileged actions, or audit evidence.

It provides:
- memory storage and recall
- provenance/context records
- trust checks before retrieved information is reused
- audit-ready memory activity

Important:
Stored ≠ trusted.
Recalled ≠ authorized for reuse.

## Search Result Trust Check API (Free)

POST /api/search-result-trust/check returns allow / deny / review_required before an AI agent relies on a retrieved result, observation, or log.

Input: agent_id / result_type / source / freshness / provenance / contradictions / intended_use / policy
Output: decision / risk_level / reason / freshness_status / provenance_status / contradiction_status / evidence_id

Use when: retrieving search results, operational logs, observations, or SOPs before tool execution, payment decisions, or physical actions.

This MVP is free, stateless, and not included in x402 manifest.

## Paid Memory Operations

Main endpoint:
POST /api/memory/store (0.05 USDC)

Input: agent_id / content / memory_type / ttl_seconds
Output: memory_id / stored / audit_log_id / next_recommended

Recommended flow:
x402 Payment -> Paid API -> Agent Memory API (store audit record)

Do not use this API as a primary database or long-term personal data store.
This API is a post-execution audit memory layer.

This API does not guarantee that recalled content is true,
fresh, permitted for downstream use, or consulted by every workflow.

Do not rely on:
- mandatory memory consultation
- truth guarantee without independent verification
- freshness guarantee without explicit freshness checks
- automatic execution permission for recalled data
- payment authorization based only on recalled memory

## Note

This API is part of Agent Control Primitives.
Agent Control Primitives is not a payment rail, wallet, custodian, or x402 facilitator.
It provides lightweight safety, budget, approval, context, and audit primitives around agent actions and payments.

## Agent Action Atom

Agent Action Atom is a minimal external record unit for one AI-agent action.

In the context of agent-memory-api, use it to record memory write decisions, memory access policy checks, boundary enforcement outcomes, and audit evidence for memory operations.

Agent Action Atom is not a payment protocol, wallet, settlement layer, legal liability engine, or API endpoint by itself.

Agent Action Atoms can be grouped into higher-level external control materials:
- Payment Trace
- Evidence Packet
- Boundary Trace
- Fuse Record
- Human Approval Trail
- Permission Record
- Dispute Record
- Liability Evidence

Current status: private design concept, external control material, minimal record unit, not an API, not a protocol, not a standard.

## Memory Provenance Context Record Builder

Memory Provenance Context Record Builder is a free experimental stateless builder that creates external control material for AI-agent memory and context usage.

Endpoint: POST /api/memory-provenance-record/build (free, no x402 required)

It can describe:
- raw sources
- extracted facts
- profile or context summary
- memory layer
- context state / use rule / evidence / last_checked
- freshness requirement
- risk flags
- Atom-compatible action reference

Use this when an AI agent needs to know whether a memory, project status, service status, or context item can be used for tool permission, spending policy, payment decision, or evidence packet workflows.

The builder is free because it creates the provenance and state record structure only.

Actual memory storage, recall, encryption, search, or verification remain handled by memory operation endpoints.

It is not a memory store, not a vector database, not a model provider, not a payment protocol, not a wallet, not a settlement layer, not a legal compliance system, and not an official standard.

## External Control Materials

This service provides the trust-side materials in the External Control Materials flow for AI agents.

It helps agents answer: Where did this memory come from? Can this retrieved result be trusted?

Central AI-readable map:
https://ai-agent-payment-safety-stack.onrender.com/.well-known/external-control-materials.json

OKF-style markdown knowledge bundle:
https://ai-agent-payment-safety-stack.onrender.com/okf/index.md

Core gates: trust (Search Result Trust Check) / execution (Tool Approval Check) / payment (Payment Review)

Memory-side local flow: Memory Provenance → Search Result Trust Check → Tool Approval → Payment Review → Evidence

Related OKF concepts:
- https://ai-agent-payment-safety-stack.onrender.com/okf/materials/search_result_trust_check.md

## Recommended buyer flow

Use agent-memory-api to add memory boundaries for AI agents.

Recommended order:

1. POST /api/memory/store
   Store memory that should be retained.

2. POST /api/memory/recall
   Recall relevant memory.

3. When recalled or retrieved information may influence
   tools, payments, privileged actions, or operational decisions:
   POST /api/search-result-trust/check

4. POST /api/context/package
   Package selected context for downstream use.

5. GET /api/memory/audit
   Review memory activity and evidence later.

Use GET /api/memory/audit.
Do not POST to /api/memory/audit.

Optional:
- POST /api/trust/verify
- POST /api/recall/compress
- POST /api/recall/extract
- POST /api/memory/delete

If you receive 402 Payment Required:

1. Read the x402 payment requirements returned by the API.
2. Pay using a compatible x402 client.
3. Retry the same request after payment.
4. Use the returned result before continuing the agent workflow.