Workspace Memory

Workspace memory is a team-scoped knowledge store for capturing decisions, retrospectives, standups, and other organizational context that all agents in a workspace should know about. It is separate from user-specific memory (Tiers 2–3) and from the knowledge base (which is for structured reference material).

What workspace memory stores

Workspace memory is designed for collaborative, team-level context:

Architecture decisions and rationale
Sprint retrospective notes and action items
Daily standup summaries
Team agreements and working norms
Project status updates
Incident post-mortems

Every entry is scoped to a workspace ID, so different teams can have isolated memory even on the same Astra instance.

Tools

Agents interact with workspace memory using two tools:

workspace_memory_write

json

{
  "tool": "workspace_memory_write",
  "params": {
    "type": "decision",
    "title": "Use pgvector for all vector search",
    "content": "We evaluated Pinecone, Weaviate, and pgvector. We chose pgvector because it eliminates a separate service dependency and our data volume (~50M vectors) is well within its performance envelope at our QPS target.",
    "tags": ["database", "vector-search", "architecture"],
    "author": "engineering-team"
  }
}

workspace_memory_search

json

{
  "tool": "workspace_memory_search",
  "params": {
    "query": "vector database decision",
    "types": ["decision", "adr"],
    "limit": 5
  }
}

Entry types

Type	Description
`decision`	A team decision with context and rationale
`adr`	Architecture Decision Record (structured format)
`standup`	Daily standup summary
`retrospective`	Sprint retrospective notes
`incident`	Incident post-mortem
`note`	General team note

Access control

Workspace memory entries are accessible to all agents running within the same workspace. Agents from different workspaces cannot access each other's workspace memory. If your deployment has multiple teams on a single instance, each team should be assigned its own workspace ID.

REST API access

Workspace memory can also be managed directly via the REST API:

bash

# Create a workspace memory entry
POST /workspaces/:id/memory
{
  "type": "decision",
  "title": "...",
  "content": "..."
}

# Search workspace memory
GET /workspaces/:id/memory/search?q=pgvector&types=decision,adr&limit=10

# List recent entries
GET /workspaces/:id/memory?type=standup&limit=20