Skip to content

Ambient context search: automatic project-aware retrieval (ADR-008) #28

@Paul-Kyle

Description

@Paul-Kyle

Problem

With many memory files across multiple projects, search results from unrelated projects drown out relevant ones. A search for "ADR-004" from a palinode working directory returns results from other projects instead of the palinode-specific memory.

Solution

Automatic ambient context boost. When searching from a project directory, memories tagged with matching project entities get a score boost after hybrid search fusion. Not a filter — a lens. Cross-project results still appear if they're genuinely the best match.

How it works

  1. Context resolution: CWD → project entity (configurable map or auto-detect from directory name)
  2. Entity boost: Results from files tagged with the matching project entity get a 1.5x score multiplier
  3. Embedding augmentation: Query is prepended with project context before embedding for semantic-level awareness
  4. CLI flag: --no-context to disable when you want unbiased results

Interfaces

  • MCP: Auto-detects from PALINODE_PROJECT env var or CWD
  • API: Optional context field on search request
  • CLI: Auto-detects from CWD, --no-context to disable

Config

context:
  enabled: true
  boost: 1.5
  auto_detect: true
  project_map:
    my-project: project/my-project

Status

Phase G1 (project entity boost) is implemented with 10 tests. See ADR-008 for full design including Phase G2 (server-side session context) and G3 (multi-signal fusion with learned weights).

No existing AI memory system (Mem0, Zep, Letta, LangMem) does automatic ambient context fusion for retrieval. This is novel.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions