CLI¶
AIMU's CLI surface is a handful of python -m entry points: the bundled FastMCP tool/memory/prompt servers, plus an A2A agent server.
Tools server¶
Runs a FastMCP server that registers every function in aimu.tools.builtin.ALL_TOOLS (weather, search, calculate, get_webpage, wikipedia, list_directory, read_file, echo, get_current_date_and_time).
Connect from another process:
from aimu.tools import MCPClient
mcp = MCPClient({"mcpServers": {"aimu": {"command": "python", "args": ["-m", "aimu.tools.mcp"]}}})
A2A agent server¶
Requires the a2a extra (pip install 'aimu[a2a]'). Exposes a single AIMU Agent over the Agent2Agent protocol: the agent-level analog of the tools server above (which serves tools; this serves a whole agent).
python -m aimu.agents.a2a \
--model anthropic:claude-sonnet-4-6 \
--system "You are a helpful research assistant." \
--port 9000
Flags: --model (defaults to AIMU_LANGUAGE_MODEL / a locally available model), --system, --name, --host (default 127.0.0.1), --port (default 9000).
Connect from another process:
from aimu.agents import RemoteAgent
remote = RemoteAgent.connect("http://localhost:9000")
print(remote.run("Summarise the latest on small language models."))
To serve an arbitrary Runner (a workflow, a custom agent) rather than a plain Agent, call aimu.agents.serve_a2a(runner, port=9000) from your own script instead of this CLI.
Semantic memory server¶
Wraps SemanticMemoryStore. Tools registered:
search_memories(search_request)add_memories(memories)delete_memory(memory)list_memories()
Reads MEMORY_STORE_PATH (default: in-memory).
Document memory server¶
Wraps DocumentStore with tools matching Anthropic's Managed Agents Memory API:
memory_list(path_prefix)memory_search(query)memory_read(path)memory_write(path, content)memory_edit(path, old_str, new_str)memory_delete(path)
Reads DOCUMENT_STORE_PATH (default: in-memory).
Prompt catalog server¶
Wraps PromptCatalog. Tools registered:
get_prompt(name, model_id)list_prompts()store_prompt_version(name, model_id, prompt, metrics)
Reads PROMPT_CATALOG_PATH (default: prompts.db in cwd).
Web chat apps¶
Not strictly CLI but launched the same way:
streamlit run web/streamlit_chatbot.py # Streamlit UI
python web/gradio_chatbot_basic.py # Gradio UI
Both demo a full-featured chat with streaming, tool calls, and persistent history.
See also¶
- Environment variables: every var these commands read
- How-to: use MCP tools: attaching MCP servers to a client