adding timestamp knowledge to agent prompt and info

[ClaireBookworm]

small fix, prompts/agent info to tell it to pay attention to graph_knowledge timestamps and receive the data

[greptile-apps]

Greptile Overview

Greptile Summary

This PR enhances the temporal memory system by adding timestamp information to entity data. The changes enable the VLM agent to answer questions about when entities appeared, disappeared, or how long they were visible.

Key changes:

• Added entity_timestamps array to graph_context containing first_seen_ts, last_seen_ts, and calculated duration_s for each entity
• Updated query prompt to instruct the agent on how to use timestamp data for duration-based questions
• One minor issue: unused time import added to graph_utils.py

The implementation correctly retrieves timestamp data from the entity database and properly calculates duration. The prompt documentation is clear and helpful for the agent.

Confidence Score: 4/5

• This PR is safe to merge with only a minor style issue
• The changes are straightforward and well-implemented. The logic for extracting and calculating entity timestamps is correct, and the prompt updates are clear. The only issue is an unused import which doesn't affect functionality.
• No files require special attention

Important Files Changed

Filename	Overview
dimos/perception/experimental/temporal_memory/temporal_utils/graph_utils.py	Added entity timestamp tracking (entity_timestamps) for temporal queries; unused time import added
dimos/perception/experimental/temporal_memory/temporal_utils/prompts.py	Updated query prompt instructions to explain entity_timestamps field and how to use it for duration queries

Sequence Diagram

sequenceDiagram
    participant User
    participant TemporalMemory
    participant GraphUtils
    participant GraphDB
    participant VLM

    User->>TemporalMemory: query(question, latest_frame)
    TemporalMemory->>GraphUtils: build_graph_context(graph_db, entity_ids)
    
    Note over GraphUtils: NEW: Build entity_timestamps array
    
    loop For each entity_id
        GraphUtils->>GraphDB: get_entity(entity_id)
        GraphDB-->>GraphUtils: entity data (first_seen_ts, last_seen_ts)
        GraphUtils->>GraphUtils: Calculate duration_s = last_seen_ts - first_seen_ts
        GraphUtils->>GraphUtils: Append to entity_timestamps[]
    end
    
    loop For each entity_id (existing logic)
        GraphUtils->>GraphDB: get_relations_for_entity()
        GraphDB-->>GraphUtils: relationships
        GraphUtils->>GraphDB: get_nearby_entities()
        GraphDB-->>GraphUtils: spatial info
        GraphUtils->>GraphDB: get_semantic_relations()
        GraphDB-->>GraphUtils: semantic knowledge
    end
    
    GraphUtils-->>TemporalMemory: graph_context (with entity_timestamps)
    TemporalMemory->>GraphUtils: build_query_prompt(question, context)
    
    Note over GraphUtils: NEW: Prompt includes instructions<br/>about entity_timestamps field
    
    GraphUtils-->>TemporalMemory: prompt (with timestamp guidance)
    TemporalMemory->>VLM: query(latest_frame, prompt)
    VLM-->>TemporalMemory: answer (can now use timestamp data)
    TemporalMemory-->>User: answer

Loading

[greptile-apps]

_{1 file reviewed, 1 comment}

_{Edit Code Review Agent Settings | Greptile}