CLI tool for exploring AI agent chat data.

curl -LsSf https://astral.sh/uv/install.sh | sh
uv sync --all-extras --dev
uv pip install git+https://github.com/arclabs561/multiscale.git
uv pip install -e ../llm-helpers

Optional: rank-fusion and rank-refine (not on PyPI):

uv pip install -e ../rank-fusion/rank-fusion-python
uv pip install -e ../rank-refine/rank-refine-python

uv run ae chats
uv run ae convo <composer_id>
uv run ae find-solution "query"
uv run ae remember "query"
uv run ae ensure-indexed

Basic:

• info - Database info
• chats - List conversations
• convo <id> - Show conversation
• keys [--prefix] [--like] - List keys
• show <key> - Show key value

Search:

• find-solution <query> - Search conversation history
• remember <query> - Recall with optional LLM summarization
• ensure-indexed - Build indexes

Analysis:

• pairs <id> - Extract QA pairs
• multiscale - Hierarchical summarization
• design-coherence - Organize design plans

See uv run agent-explorer --help for all commands.

• AGENT_TYPE - Agent type (default: cursor)
• AGENT_STATE_DB / CURSOR_STATE_DB - Database path
• AGENT_INDEX_JSONL / CURSOR_INDEX_JSONL - Index path (default: ./cursor_index.jsonl)
• AGENT_VEC_DB / CURSOR_VEC_DB - Vector DB path (default: ./cursor_vec.db)
• OPENAI_API_KEY - Required for LLM features
• OPENAI_MODEL - Model (default: gpt-4o-mini)
• OPENAI_EMBED_MODEL - Embedding model (default: text-embedding-3-small)

• Agent-agnostic design (Cursor implemented, extensible for Cline/Aider)
• Multi-source search (vector + sparse, rank-fusion optional)
• Conversation recall (LLM summarization optional)
• Key-value access to agent databases
• Hierarchical summarization
• Idempotent indexing with caching
• Read-only database access

• Commands output JSON (pipe to jq for formatting)
• Close agent before accessing database