version: "1.0.0" title: AgenticFleet Memory System Guide tags: [memory, documentation, setup] author: AgenticFleet Architect created: 2024-12-29

AgenticFleet Memory System Guide

A comprehensive guide to using the Two-Tier Memory System (Local + Chroma Cloud) for persistent agent context.

Context

When to use this skill:

• You are a new agent or contributor starting a session.
• You need to recall past architectural decisions or debugging solutions.
• You want to save a new complex solution for the future.

Prerequisites:

• Python dependencies installed: uv pip install -r .context/scripts/requirements.txt
• Chroma Cloud configured in .context/.chroma/config.yaml.

Step-by-Step Guide

1. Initialization

At the start of a session, always hydrate the local context.

uv run python .fleet/context/scripts/memory_manager.py init

2. Recalling Information (Semantic Search)

To find information stored in Chroma Cloud:

uv run python .fleet/context/scripts/memory_manager.py recall "how to fix cors error"

This queries the semantic (facts) and procedural (skills) collections.

3. Learning (Saving New Skills)

When you solve a tough problem:

1. Create a markdown file in .context/skills/ (e.g., fix-auth-bug.md).
2. Use the SKILL_TEMPLATE.md structure.
3. Index it into Chroma:

``bash uv run python .context/scripts/memory_manager.py learn --file .context/skills/fix-auth-bug.md ``

4. Reflection (End of Session)

Summarize your work to keep the history clean.

uv run python .fleet/context/scripts/memory_manager.py reflect

Common Pitfalls

• Not Configuring Chroma: If config.yaml is missing or has no keys, the system falls back to local-only (no search).
• Committing Private Data: Never force-add files in core/ or skills/ unless they are generic templates.
• Forgetting to Learn: Solving a problem without running /learn means the next agent has to solve it again.

Related Resources

• AGENTS.md (System Instructions)
• .fleet/context/MEMORY.md (Architecture Docs)