TrueMemory/Documentation

TrueMemory / Human + agent docs

Understand the memory layer. Build on it.

Product story, architecture reference, developer guide, and agent-readable source of truth — built from the repository’s verified implementation.

Quickstart · the memory loop

One token, one Space, three calls. The same flow works for an application or an agent.

const memory = await fetch("/v1/memories", {
  method: "POST", headers: auth, body: JSON.stringify({ content })
});
await fetch("/v1/memories/search", { method: "POST", headers: auth, body: JSON.stringify({ query }) });
await fetch("/v1/memories/retrieve", { method: "POST", headers: auth, body: JSON.stringify({ memory_ids }) });
Read the REST contract

Start here

Learn the product model and make your first memory.

Build with TrueMemory

REST, TypeScript, Python, HTTP, and MCP integration paths.

Architecture

See how clients, authorization, memory, retrieval, and storage connect.

Security

Scopes, bindings, isolation, token lifecycle, and failure boundaries.

Start here

What is TrueMemory?

Universal memory infrastructure for agents, applications, models, and sessions.

TrueMemory is the memory layer. It exposes a REST Memory API, MCP, and SDK-friendly HTTP interfaces over scoped storage and retrieval. TrueMemory Assistant is a first-party client, not the platform identity.

Start here

First memory

Store, search, and retrieve a scoped memory in minutes.

Create a scoped token with the memory scope, POST a memory to /v1/memories, then call /v1/memories/search or /v1/memories/retrieve. Bind tokens to a Space or agent when isolation requires it.

Architecture

System overview

REST, MCP, and SDK clients converge on the memory boundary.

Clients authenticate at the API boundary. The REST and MCP routes authorize scopes and bindings before delegating to the memory client. Storage, retrieval, temporal filtering, rate limiting, and observability remain behind that boundary.

Memory

Memory lifecycle

Ingest → validate → store → retrieve → revise → forget.

The current implementation supports memory writes, search, retrieval, update, forgetting, workspace and agent bindings, temporal parameters, provenance, confidence, and history-aware retrieval where the underlying memory data provides it.

Developers

REST API

The canonical HTTP contract for memory infrastructure.

GET /v1/memories lists authorized memories. POST /v1/memories writes. POST /v1/memories/search and POST /v1/memories/retrieve recall context. POST /v1/memories/update and POST /v1/memories/forget mutate it. Authentication requires a token with the memory scope.

Developers

MCP

Connect an MCP-compatible agent to TrueMemory.

The MCP JSON-RPC endpoint is POST /mcp. Available memory tools include memory_search, memory_retrieve, memory_context, memory_entities, memory_remember, and memory_update, subject to authentication, scope, bindings, and rate limits.

Security

Isolation and authorization

Tenant, Space, and agent bindings are enforced at the boundary.

Authorization is enforced before memory operations. Token bindings for workspace_id and agent_id cannot be widened by a client request. Disallowed bindings return an authorization error rather than crossing the boundary.

History

What is verified?

The repository is the historian; gaps are not filled with invented biography.

The repository contains earlier TrueMemory, AmanAgentLab, and AmanCrawl vision documents as historical source material. The exact chronology of every experiment and the founder's personal motivations are not fully verified here and are intentionally marked unknown.