19 KiB
Structured Memory
See also: Memory Architecture diagram
The structured memory system gives the AI persistent, typed, scoped memories that survive across sessions and workstreams. Memories are automatically surfaced in the system message via BM25 relevance scoring, so the model has contextual recall without explicit search.
Overview
Each memory has three dimensions:
- Type -- categorizes the memory's purpose
- Scope -- controls visibility boundaries
- Name -- unique identifier within a scope (snake_case, normalized)
Memory types
| Type | Purpose |
|---|---|
user |
User preferences, conventions, working style |
general |
General knowledge, architecture, patterns |
feedback |
Corrections, lessons learned, things to avoid |
reference |
Reference material, documentation, specifications |
Memory scopes
| Scope | Visibility |
|---|---|
global |
Visible to all workstreams and users |
workstream |
Visible only within the originating workstream |
user |
Follows the authenticated user across workstreams |
coordinator |
Coordinator sessions only; follows the acting user |
project |
Shared by workstreams attached to one active project |
A memory's identity is the tuple (name, scope, scope_id). Saving a memory
with the same identity upserts -- updating content while preserving the ID.
Inherited target and coordinator scope
Name-based operations use one inherited target when scope is omitted:
- An attached active project selects
projectforsave,get, anddelete. - Read-only project access permits
get, butsaveanddeletefail. They do not fall back to a broader namespace. - Without a project, interactive sessions select
global; coordinator sessions selectcoordinator.
A valid explicit scope selects exactly that scope. search and list are the
only actions that span every visible scope when scope is omitted.
Each coordinator's private coordinator namespace is keyed by the acting
user's user_id. It is durable -- every coordinator session that user runs
(including concurrent ones) shares one orchestration namespace, so procedures
and lessons survive close/reopen.
Isolation is bidirectional and enforced by session kind, not by secrecy of the scope id:
- A coordinator session sees its acting user's
coordinatorscope and, when attached, the sharedprojectscope. It never seesglobal/workstream/usermemories. - Interactive sessions -- including a coordinator's own children, which share
its
user_id-- are rejected from thecoordinatorscope on every memory action. Children cannot plant rows the parent coordinator would read. - The REST memory API (
/v1/api/memories) does not accept thecoordinatorscope at all; the scope is written exclusively through a coordinator session's own memory tool.
Coordinator sessions require an authenticated user identity -- an anonymous coordinator cannot be constructed, so the scope id is always a real user.
BM25 relevance injection
On every conversation turn, the system:
- Resolves the acting principal and their live project access
- Fetches up to
fetch_limitmemories across that visibility envelope - Extracts context from the last 3 user messages
- Scores memories against that context using a BM25 index
- Injects the top
relevance_kmemories into the system message as<memories>XML tags - Appends a hint telling the model how many memories are in scope
This means the model always has its most relevant memories available without
explicit recall -- but can still use memory(action='search') for deeper
lookup.
The persona memory lever gates this pathway: a workstream whose persona turns memory off receives no relevance injection at all -- the steps above run only when memory is enabled for the session. See Personas.
Nudges
The metacognition layer can nudge the model to save memories at appropriate
moments (e.g., after a correction or when resuming a workstream). Nudges are
rate-limited by nudge_cooldown and can be disabled entirely.
Configuration
config.toml
[memory]
relevance_k = 5 # top-k memories injected per turn
fetch_limit = 50 # max memories fetched from storage for scoring
max_content = 32768 # max content length per memory (characters)
nudge_cooldown = 300 # minimum seconds between memory nudges
nudges = true # enable/disable metacognitive nudges
All fields are optional. Defaults are shown above.
Tool Usage
The memory tool supports five actions:
save
Store or update a memory.
Every save is a complete write for the relevance summary: description must
be supplied and contain non-whitespace text on both creation and update.
Content-only updates are rejected.
{
"action": "save",
"name": "project_architecture",
"content": "The project uses a hexagonal architecture with...",
"description": "Core architecture patterns",
"type": "general",
"scope": "global"
}
| Parameter | Required | Default | Description |
|---|---|---|---|
name |
yes | -- | Snake_case identifier (max 256 chars) |
content |
yes | -- | Memory content (max max_content chars) |
description |
yes | -- | Non-empty relevance summary, required on create and update |
type |
no | "general" |
One of: user, general, feedback, reference |
scope |
no | inherited | Kind-valid scope; see inherited target above |
get
Retrieve the full content of one memory by name.
{
"action": "get",
"name": "project_architecture",
"scope": "project"
}
| Parameter | Required | Default | Description |
|---|---|---|---|
name |
yes | -- | Memory name to retrieve |
scope |
no | inherited | Exact scope to query |
search
Find memories by query (BM25 full-text search).
{
"action": "search",
"query": "authentication patterns",
"type": "general",
"limit": 10
}
| Parameter | Required | Default | Description |
|---|---|---|---|
query |
yes | -- | Search query |
type |
no | "" |
Filter by type |
scope |
no | "" |
Filter by scope |
limit |
no | 20 |
Max results (capped at 50) |
delete
Remove a memory by name.
{
"action": "delete",
"name": "outdated_pattern",
"scope": "global"
}
| Parameter | Required | Default | Description |
|---|---|---|---|
name |
yes | -- | Memory name to delete |
scope |
no | inherited | Exact scope to delete |
list
List all memories with optional filters.
{
"action": "list",
"type": "feedback",
"limit": 50
}
| Parameter | Required | Default | Description |
|---|---|---|---|
type |
no | "" |
Filter by type |
scope |
no | "" |
Filter by scope |
limit |
no | 20 |
Max results (capped at 50) |
Server API
Four endpoints on the server for programmatic memory access.
GET /v1/api/memories
List memories with optional filters.
Without scope, the response is restricted to global plus the authenticated
caller's user namespace. The public API accepts only global, user, and
workstream; internal project and coordinator namespaces remain available
through the session tool and admin API. Explicit workstream access requires
its persisted owner (or a service token).
Query parameters:
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
type |
string | no | "" |
Filter by memory type |
scope |
string | no | "" |
Filter by scope |
scope_id |
string | no | "" |
Filter by scope ID |
limit |
int | no | 100 |
Max results (1-200) |
When scope=user, the authenticated user's ID is used automatically and a
different supplied ID is rejected. scope=workstream requires scope_id.
Response: 200
{
"memories": [
{
"memory_id": "a1b2c3d4-e5f6-...",
"name": "project_architecture",
"description": "Core architecture patterns",
"type": "general",
"scope": "global",
"scope_id": "",
"content": "The project uses a hexagonal architecture...",
"created": "2026-03-10T10:00:00",
"updated": "2026-03-12T14:30:00"
}
],
"total": 1
}
POST /v1/api/memories
Save or upsert a structured memory.
description is mandatory for both creates and updates and must contain
non-whitespace text. The API rejects content-only updates.
Request body:
{
"name": "deployment_process",
"content": "Deploy via GitHub Actions. Staging auto-deploys on push to main.",
"description": "CI/CD deployment workflow",
"type": "general",
"scope": "global",
"scope_id": ""
}
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
name |
string | yes | -- | Memory name (max 256 chars) |
content |
string | yes | -- | Memory content (max 65536 chars) |
description |
string | yes | -- | Non-empty relevance summary, required on create and update |
type |
string | no | unset | user, general, feedback, or reference |
scope |
string | no | "global" |
One of: global, workstream, user |
scope_id |
string | no | "" |
Scope qualifier (auto-resolved for user scope) |
Response (created): 201
{
"memory_id": "a1b2c3d4-e5f6-...",
"name": "deployment_process",
"description": "CI/CD deployment workflow",
"type": "general",
"scope": "global",
"scope_id": "",
"content": "Deploy via GitHub Actions...",
"created": "2026-03-14T10:00:00",
"updated": "2026-03-14T10:00:00"
}
Response (updated): 200 -- same schema, returned when a memory with the
same (name, scope, scope_id) already existed.
Errors:
| Status | Condition |
|---|---|
| 400 | Invalid input, scope, scope ID, or limit |
| 403 | Cross-user or non-owner workstream access |
| 404 | Explicit workstream does not exist |
| 500 | Storage mutation failed |
POST /v1/api/memories/search
Search memories by query. Uses POST for the request body but is non-mutating
(requires only read scope).
An omitted scope searches the same caller-bound global + user envelope as
the list endpoint. It never means every row in the table.
Request body:
{
"query": "authentication",
"type": "general",
"scope": "",
"scope_id": "",
"limit": 20
}
| Field | Type | Required | Default | Description |
|---|---|---|---|---|
query |
string | yes | -- | Search query |
type |
string | no | "" |
Filter by type |
scope |
string | no | "" |
Filter by scope |
scope_id |
string | no | "" |
Filter by scope ID |
limit |
int | no | 20 |
Max results (1-50) |
Response: 200
{
"memories": [
{
"memory_id": "a1b2c3d4-e5f6-...",
"name": "auth_patterns",
"description": "Authentication architecture",
"type": "general",
"scope": "global",
"scope_id": "",
"content": "JWT tokens with HS256...",
"created": "2026-03-10T10:00:00",
"updated": "2026-03-12T14:30:00"
}
],
"total": 1
}
DELETE /v1/api/memories/{name}
Delete a memory by name and scope.
Deletes are atomic: the row used for the success result and audit event is the
row actually removed. A storage failure returns 500, not a false 404.
Path parameters:
| Parameter | Type | Description |
|---|---|---|
name |
string | Memory name |
Query parameters:
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
scope |
string | no | "global" |
Scope of the memory |
scope_id |
string | no | "" |
Scope qualifier |
Response (success): 200
{"status": "ok", "name": "deployment_process"}
Response (not found): 404
{"error": "Memory 'deployment_process' not found"}
Console Admin API
Four admin endpoints for cross-workstream memory management. All require the
admin.memories permission.
GET /v1/api/admin/memories
List memories across all scopes (no automatic scope resolution).
Query parameters:
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
type |
string | no | "" |
Filter by type |
scope |
string | no | "" |
Filter by scope |
scope_id |
string | no | "" |
Filter by scope ID |
limit |
int | no | 100 |
Max results (capped at 200) |
Response: 200
{
"memories": [
{
"memory_id": "a1b2c3d4-e5f6-...",
"name": "project_architecture",
"description": "Core architecture patterns",
"type": "general",
"scope": "global",
"scope_id": "",
"content": "The project uses...",
"created": "2026-03-10T10:00:00",
"updated": "2026-03-12T14:30:00"
}
],
"total": 1
}
GET /v1/api/admin/memories/search
Search memories by query (uses query parameters, not POST body).
Query parameters:
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
q |
string | yes | -- | Search query |
type |
string | no | "" |
Filter by type |
scope |
string | no | "" |
Filter by scope |
scope_id |
string | no | "" |
Filter by scope ID |
limit |
int | no | 20 |
Max results (capped at 50) |
Response: 200 -- same schema as GET /v1/api/admin/memories.
GET /v1/api/admin/memories/{memory_id}
Get a single memory by ID.
Path parameters:
| Parameter | Type | Description |
|---|---|---|
memory_id |
string | Memory UUID |
Response (success): 200
{
"memory_id": "a1b2c3d4-e5f6-...",
"name": "project_architecture",
"description": "Core architecture patterns",
"type": "general",
"scope": "global",
"scope_id": "",
"content": "The project uses...",
"created": "2026-03-10T10:00:00",
"updated": "2026-03-12T14:30:00"
}
Response (not found): 404
{"error": "Memory not found"}
DELETE /v1/api/admin/memories/{memory_id}
Delete a memory by ID. Records an audit event (memory.delete).
Path parameters:
| Parameter | Type | Description |
|---|---|---|
memory_id |
string | Memory UUID |
Response (success): 200
{"status": "ok"}
Response (not found): 404
{"error": "Memory not found"}
SDK
Python
The server SDK uses mem_type (not type) to avoid shadowing the Python
builtin.
from turnstone.sdk import TurnstoneServer
with TurnstoneServer("http://localhost:8080", token="tok_xxx") as client:
# Save a memory
mem = client.save_memory(
"api_conventions",
"All endpoints use /v1/ prefix. JSON responses.",
description="API design patterns",
mem_type="general",
scope="global",
)
print(mem.memory_id)
# Search memories
results = client.search_memories("authentication", mem_type="general", limit=10)
for m in results.memories:
print(f"{m['name']}: {m['description']}")
# List memories
all_mems = client.list_memories(mem_type="feedback", limit=50)
# Delete a memory
client.delete_memory("api_conventions", scope="global")
Console admin SDK:
from turnstone.sdk import TurnstoneConsole
with TurnstoneConsole("http://localhost:9090", token="tok_xxx") as admin:
# List all memories (admin view, no scope auto-resolution)
result = admin.list_memories(scope="global", limit=100)
# Search
result = admin.search_memories("architecture", mem_type="general")
# Get by ID
mem = admin.get_memory("a1b2c3d4-e5f6-...")
# Delete by ID
admin.delete_memory("a1b2c3d4-e5f6-...")
TypeScript
import { TurnstoneServer } from "@turnstone/sdk";
const client = new TurnstoneServer({
baseUrl: "http://localhost:8080",
token: "tok_xxx",
});
// Save a memory
const mem = await client.saveMemory({
name: "api_conventions",
content: "All endpoints use /v1/ prefix. JSON responses.",
description: "API design patterns",
type: "general",
scope: "global",
});
// Search memories
const results = await client.searchMemories({
query: "authentication",
type: "general",
limit: 10,
});
// List memories
const all = await client.listMemories({ type: "feedback", limit: 50 });
// Delete a memory
await client.deleteMemory("api_conventions", { scope: "global" });
Console admin SDK:
import { TurnstoneConsole } from "@turnstone/sdk";
const admin = new TurnstoneConsole({
baseUrl: "http://localhost:9090",
token: "tok_xxx",
});
// List, search, get, delete by ID
const mems = await admin.listMemories({ scope: "global" });
const found = await admin.searchMemories({ q: "auth", limit: 20 });
const one = await admin.getMemory("a1b2c3d4-e5f6-...");
await admin.deleteMemory("a1b2c3d4-e5f6-...");
Storage
Memories are stored in the structured_memories table (migration 013).
The unique constraint on (name, scope, scope_id) ensures upsert semantics.
The name is normalized on save: lowercased, hyphens and spaces replaced with
underscores.
Architecture
See Memory Architecture diagram for the full data flow covering the session tool path, API path, admin path, and BM25 relevance injection.