Memory & Knowledge

Astonish has a persistent memory system that lets the agent remember facts, preferences, and project details across sessions. It is powered by a local vector store (chromem-go) with local embeddings — no external API is needed.

How it works

The agent automatically retrieves relevant knowledge before responding. If you have told it something before, it remembers. No manual searching is needed on your part.

Before each response, Astonish searches memory for content relevant to your current question and injects matching context into the system prompt.

Memory tools

The agent has three tools for working with memory:

Tool	Description
memory_save	Store facts, preferences, and infrastructure details. Content is organized by category headings in markdown files.
memory_search	Semantic search across all indexed memory. Returns scored snippets with file references.
memory_get	Read specific lines from a memory file for full context.

Memory files

Memory files are stored in ~/.config/astonish/memory/ as markdown files.

• MEMORY.md — the default file for core identity and preferences.
• Topic-specific files — knowledge about particular subjects goes in separate files (e.g., infra/portainer.md, projects/myapp.md).
• SELF.md — the agent’s self-awareness file containing its identity, preferences, and communication style. Auto-indexed.
• INSTRUCTIONS.md — per-project instructions. Place this in a project directory and Astonish reads it when working in that directory.

CLI commands

Search memory from the terminal:

astonish memory search <query>

List memory files and chunk counts:

astonish memory list

Show memory system status:

astonish memory status

Force re-index all memory files:

astonish memory reindex

Configuration

These settings can be adjusted through Studio Settings > Memory, or in config.yaml:

memory:
  enabled: true
  memory_dir: ~/.config/astonish/memory
  vector_dir: ~/.config/astonish/vectors
  embedding:
    # provider and model settings
  chunking:
    # chunk size and overlap params
  search:
    max_results: 10
    min_score: 0.5