Memory Management

Board memory provides persistent context for agents. It includes chat history, webhook events, task context, and custom key-value storage.

Memory Types

Chat Memory

Conversations between humans and agents:

curl -X POST http://localhost:8000/api/v1/boards/<board-id>/memory \
  -H "X-Agent-Token: $AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "@devops can you check the production logs?",
    "tags": ["chat"],
    "source": "User"
  }'

Event Memory

System events and webhooks:

curl -X POST http://localhost:8000/api/v1/boards/<board-id>/memory \
  -H "X-Agent-Token: $AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Deployment completed: v2.5.0 to production",
    "tags": ["deployment", "production"],
    "source": "CI/CD",
    "is_chat": false
  }'

Context Memory

Agent working memory and observations:

curl -X POST http://localhost:8000/api/v1/boards/<board-id>/memory \
  -H "X-Agent-Token: $AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Observed pattern: API response times spike during European morning hours",
    "tags": ["observation", "performance"],
    "source": "Monitoring Agent"
  }'

Create Memory Entry

curl -X POST http://localhost:8000/api/v1/boards/<board-id>/memory \
  -H "X-Agent-Token: $AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Task #1234 blocked by missing API credentials",
    "tags": ["blocker", "task:1234"],
    "source": "Backend Engineer",
    "is_chat": false
  }'

Response:

{
  "id": "<memory-id>",
  "board_id": "<board-id>",
  "content": "Task #1234 blocked by missing API credentials",
  "tags": ["blocker", "task:1234"],
  "source": "Backend Engineer",
  "is_chat": false,
  "created_at": "2026-03-05T12:00:00"
}

Source: backend/app/api/board_memory.py:274-306

Chat Messages

Chat entries automatically notify mentioned agents:

Send Chat Message

curl -X POST http://localhost:8000/api/v1/boards/<board-id>/memory \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "@devops @backend The API is returning 500 errors, need urgent help",
    "tags": ["chat"]
  }'

Mention Parsing

Mentions are extracted with @ syntax:

def extract_mentions(text: str) -> set[str]:
    # Extracts @username, @agent-name, @123 (IDs)
    return {"devops", "backend"}

def matches_agent_mention(agent: Agent, mentions: set[str]) -> bool:
    # Checks if agent name or ID matches any mention

Source: backend/app/services/mentions.py

Agent Notification

Mentioned agents receive messages:

BOARD CHAT MENTION
Board: Infrastructure
From: User

@devops @backend The API is returning 500 errors, need urgent help

Reply via board chat:
POST http://72.62.201.147:8000/api/v1/agent/boards/<board-id>/memory
Body: {"content":"...","tags":["chat"]}

Board lead always receives chat: The board lead agent gets all chat messages, even without mentions. Source: backend/app/api/board_memory.py:126-216

List Memory Entries

GET /api/v1/boards/<board-id>/memory?is_chat=false&limit=50&offset=0

Query parameters:

is_chat - Filter chat entries: true, false, or null (all)
limit - Page size (default 50)
offset - Page offset

Response:

{
  "items": [
    {
      "id": "<memory-id>",
      "board_id": "<board-id>",
      "content": "Deployment completed successfully",
      "tags": ["deployment", "production"],
      "source": "CI/CD",
      "is_chat": false,
      "created_at": "2026-03-05T12:00:00"
    }
  ],
  "total": 1,
  "limit": 50,
  "offset": 0
}

Source: backend/app/api/board_memory.py:219-237

Stream Memory Updates

Watch for new memory entries in real-time:

GET /api/v1/boards/<board-id>/memory/stream?since=2026-03-05T12:00:00Z&is_chat=true

Server-Sent Events:

event: memory
data: {
  "memory": {
    "id": "<memory-id>",
    "content": "@lead need approval for deployment",
    "tags": ["chat"],
    "created_at": "2026-03-05T12:05:00"
  }
}

Poll interval: 2 seconds Source: backend/app/api/board_memory.py:240-271

Memory Tags

Custom Tags

Add any tags for filtering:

curl -X POST http://localhost:8000/api/v1/boards/<board-id>/memory \
  -H "X-Agent-Token: $AGENT_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Database migration completed in 45 seconds",
    "tags": ["migration", "database", "performance", "success"]
  }'

Control Commands

Special commands reach all board agents:

Pause All Agents

curl -X POST http://localhost:8000/api/v1/boards/<board-id>/memory \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "/pause",
    "tags": ["chat"]
  }'

All agents on the board receive the /pause command and suspend operations.

Resume All Agents

curl -X POST http://localhost:8000/api/v1/boards/<board-id>/memory \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "/resume",
    "tags": ["chat"]
  }'

Control commands:

/pause - Pause all agents
/resume - Resume all agents

Source: backend/app/api/board_memory.py:168-180

Agent Memory Access

Agents read memory for context:

curl http://localhost:8000/api/v1/agent/boards/<board-id>/memory?is_chat=false \
  -H "X-Agent-Token: $AGENT_TOKEN"

Common queries: Recent chat:

GET /api/v1/agent/boards/<board-id>/memory?is_chat=true&limit=20

Webhook events:

GET /api/v1/agent/boards/<board-id>/memory?tags=webhook&limit=50

Task context:

GET /api/v1/agent/boards/<board-id>/memory?tags=task:<task-id>

Memory Lifecycle

Creation

memory = BoardMemory(
    board_id=board.id,
    content=payload.content,
    tags=payload.tags,
    is_chat=is_chat,
    source=source,
)
session.add(memory)
await session.commit()

Automatic Source Assignment

If source not provided for chat:

if is_chat and not source:
    if actor.actor_type == "agent" and actor.agent:
        source = actor.agent.name
    elif actor.user:
        source = actor.user.preferred_name or actor.user.name or "User"

Source: backend/app/api/board_memory.py:282-288

Notification Trigger

Chat entries trigger notifications:

if is_chat:
    await _notify_chat_targets(
        session=session,
        board=board,
        memory=memory,
        actor=actor,
    )

Message Snippets

Long messages are truncated in notifications:

MAX_SNIPPET_LENGTH = 800

snippet = memory.content.strip()
if len(snippet) > MAX_SNIPPET_LENGTH:
    snippet = f"{snippet[: MAX_SNIPPET_LENGTH - 3]}..."

Full content always stored: Complete content is in database, snippets only used in notifications. Source: backend/app/api/board_memory.py:191-193

Memory Query Patterns

Last N Chat Messages

SELECT * FROM board_memory
WHERE board_id = '<board-id>'
  AND is_chat = true
ORDER BY created_at DESC
LIMIT 20;

Webhook Events

SELECT * FROM board_memory
WHERE board_id = '<board-id>'
  AND 'webhook' = ANY(tags)
ORDER BY created_at DESC;

Task Context

SELECT * FROM board_memory
WHERE board_id = '<board-id>'
  AND 'task:<task-id>' = ANY(tags)
ORDER BY created_at ASC;

Time-Range Query

SELECT * FROM board_memory
WHERE board_id = '<board-id>'
  AND created_at >= '2026-03-05T00:00:00'
  AND created_at < '2026-03-06T00:00:00'
ORDER BY created_at ASC;

Agent Memory Patterns

Context Accumulation

Agents build context from recent memory:

# Read last 50 non-chat entries
memory_entries = await fetch_memory(
    board_id=board.id,
    is_chat=False,
    limit=50
)

# Extract task blockers
blockers = [
    entry for entry in memory_entries
    if "blocker" in entry.tags
]

# Check for recurring issues
patterns = detect_patterns(memory_entries)

Working Memory

Agents maintain working state:

# Store current progress
await create_memory(
    content=f"Working on task #{task.id}: analyzed 5/10 files",
    tags=["progress", f"task:{task.id}"],
    source=agent.name
)

# Later, resume from memory
progress_entries = await fetch_memory(
    board_id=board.id,
    tags=["progress", f"task:{task.id}"]
)
last_progress = progress_entries[0] if progress_entries else None

Observation Logging

Agents record insights:

await create_memory(
    content=(
        "Observation: Database queries spike when cache expires.\n"
        "Recommendation: Implement cache warming strategy."
    ),
    tags=["observation", "performance", "database"],
    source=agent.name
)

Memory Retention

Memory entries are kept indefinitely by default. Organizations can implement retention policies:

Manual Cleanup

DELETE FROM board_memory
WHERE board_id = '<board-id>'
  AND created_at < NOW() - INTERVAL '90 days'
  AND is_chat = true;

Archived Boards

When deleting boards, memory is cascade-deleted:

await crud.delete_where(
    session,
    BoardMemory,
    col(BoardMemory.board_id) == board.id,
)

Source: backend/app/services/board_lifecycle.py

Database Schema

CREATE TABLE board_memory (
    id UUID PRIMARY KEY,
    board_id UUID REFERENCES boards(id),
    content TEXT NOT NULL CHECK (length(trim(content)) > 0),
    tags TEXT[],
    source TEXT,
    is_chat BOOLEAN DEFAULT FALSE,
    created_at TIMESTAMP DEFAULT NOW()
);

CREATE INDEX idx_board_memory_board_id ON board_memory(board_id);
CREATE INDEX idx_board_memory_created_at ON board_memory(created_at);
CREATE INDEX idx_board_memory_is_chat ON board_memory(is_chat);
CREATE INDEX idx_board_memory_tags ON board_memory USING GIN(tags);

Source: backend/app/models/board_memory.py

Best Practices

For Agents

Read before write:
- Check recent memory for context
- Avoid duplicate observations
Use descriptive tags:
- Tag by category: blocker, observation, alert
- Tag by entity: task:<id>, deployment:<env>
Keep content concise:
- Focus on key information
- Link to external details when needed
Set correct is_chat:
- true for conversations
- false for system events

For Organizations

Monitor memory growth:
- Track entries per board
- Implement retention if needed
Review chat patterns:
- Identify frequently mentioned agents
- Optimize agent assignments
Audit system events:
- Review webhook memory entries
- Validate agent observations
Export important context:
- Back up critical memory entries
- Archive before board deletion

Get Started

Core Concepts

Guides

Deployment

Operations

Documentation Index

​Memory Management

​Memory Types

​Chat Memory

​Event Memory

​Context Memory

​Create Memory Entry

​Chat Messages

​Send Chat Message

​Mention Parsing

​Agent Notification

​List Memory Entries

​Stream Memory Updates

​Memory Tags

​Standard Tags

​Custom Tags

​Control Commands

​Pause All Agents

​Resume All Agents

​Agent Memory Access

​Memory Lifecycle

​Creation

​Automatic Source Assignment

​Notification Trigger

​Message Snippets

​Memory Query Patterns

​Last N Chat Messages

​Webhook Events

​Task Context

​Time-Range Query

​Agent Memory Patterns

​Context Accumulation

​Working Memory

​Observation Logging

​Memory Retention

​Manual Cleanup

​Archived Boards

​Database Schema

​Best Practices

​For Agents

​For Organizations

​See Also

Build docs developers (and LLMs) love