moorcheh-ai/memanto

Problem Every recall is scoped to a single active agent. Users who run multiple agents (e.g. one per project, or per customer) cannot search across them in one query. Proposed solution Route - POST /api/v2/recall/multi accepting a list of agent_ids plus the usual recall parameters (query, limit, type, min_confidence). Reads through MemoryReadService (memanto/app/services/memory_read_service.py), Moorcheh support multiple namespace search by passing list of agent namespace. Auth - decide and document the auth model: either introduce a multi-agent token or some other approach. Call this out explicitly in the PR. Client + CLI - memanto recall "<query>" --agents agent-a,agent-b,agent-c. Acceptance criteria A single query returns merged, score-sorted results across the listed agents. Each result indicates which agent it came from. Requesting an agent the caller isn't authorized for is rejected cleanly. Existing single-agent recall behavior is unchanged.

Problem Need to wire up operation to supersede an old memory with a corrected version while preserving history. The models exist : SupersedeRequest (memanto/app/models/__init__.py ~line 90) and MemorySupersedeRequest / MemorySupersedeResponse (memanto/app/models/universal_endpoints.py). Proposed solution When new information replaces an existing memory: Mark the old memory as superseded (keep it stored, flag it inactive, and link it to its successor via metadata such as superseded_by / supersedes). Store the new memory as the current version. By default recall returns only current (non-superseded) versions; add --include-superseded to walk history. Wiring: Service - supersede logic in MemoryWriteService (memanto/app/services/memory_write_service.py); read-side filtering in memanto/app/services/memory_read_service.py. Route - POST /api/v2/agents/{agent_id}/memories/{memory_id}/supersede. Client + CLI - memanto supersede <old_id> "<new content>" [--type ...] [--reason ...]. Acceptanc