daimon-client

Python client for Daimon — a pluggable AI sidecar runtime.

Installation

pip install daimon-client

Quick start

from daimon_client import Client

with Client(base_url="http://localhost:3500") as c:
    # Single LLM configured — no name needed
    reply = c.llm().chat("What is a daimon?")
    print(reply)

    # Multiple LLMs configured — pick by name
    reply = c.llm("claude").chat("What is a daimon?")

Async

import asyncio
from daimon_client import AsyncClient

async def main():
    async with AsyncClient(base_url="http://localhost:3500") as c:
        reply = await c.llm().chat("What is a daimon?")
        print(reply)

asyncio.run(main())

Streaming

with Client() as c:
    for text in c.llm().stream("Tell me a story."):
        print(text, end="", flush=True)

Multi-turn conversations

Pass a list of messages to carry history yourself:

reply = c.llm().chat([
    {"role": "user",      "content": "My name is Alice."},
    {"role": "assistant", "content": "Nice to meet you, Alice!"},
    {"role": "user",      "content": "What is my name?"},
])

Sessions

Let the sidecar maintain history server-side with a session_id:

llm = c.llm()
llm.chat("My favourite colour is blue.", session_id="chat-1")
reply = llm.chat("What is my favourite colour?", session_id="chat-1")
# reply contains "blue"

llm.clear_session("chat-1")

Inference parameters

All sampling parameters are optional and fall back to the component's configured defaults:

reply = c.llm("gpt4o").chat("Summarise this.", model="gpt-4o", temperature=0.2, max_tokens=256)

Vector store (memory)

Read and write documents in a configured vector store:

mem = c.memory("my-store")

# Upsert a document (returns the assigned ID)
doc_id = mem.upsert("The Eiffel Tower is 330 metres tall.", id="eiffel", metadata={"source": "wikipedia"})

# Semantic search
results = mem.query("tall Paris structures", top_k=3)
for r in results:
    print(f"{r.score:.3f}  {r.content}")

# Delete
mem.delete("eiffel")

Async vector store

async with AsyncClient() as c:
    mem = c.memory("my-store")
    await mem.upsert("Some content")
    results = await mem.query("my query")

Graph store

Interact with a configured graph database using Cypher:

kg = c.graph("knowledge-graph")

# Add nodes
kg.add_node(id="alice", labels=["Person"], props={"name": "Alice", "age": 30})
kg.add_node(id="bob",   labels=["Person"], props={"name": "Bob"})

# Add a relationship
kg.add_edge("alice", "bob", "KNOWS", props={"since": "2020"})

# Run a Cypher query
rows = kg.cypher(
    "MATCH (a:Person)-[:KNOWS]->(b) RETURN a.name AS from, b.name AS to"
)
print(rows)  # [{"from": "Alice", "to": "Bob"}]

# Delete a node (and all its relationships)
kg.delete_node("alice")

API reference

Client(base_url?, timeout?)

Use as a context manager (with Client() as c) or call c.close() manually.

c.llm(component="default") → LLMClient

Returns a client scoped to the named LLM component. Omit component to use whichever single LLM is configured.

prompt can be a str or a list of {"role": ..., "content": ...} dicts.

Shorthand methods on Client

c.chat(component, prompt, **kwargs), c.stream(...), c.converse(...), and c.clear_session(...) are convenience wrappers that call c.llm(component).*. They exist for quick scripts; prefer the llm() accessor for anything beyond a one-liner.

AsyncClient exposes the same API with async def methods and AsyncLLMClient via c.llm().

c.memory(store="default") → MemoryStoreClient

Returns a client scoped to the named vector store.

c.graph(store) → GraphStoreClient

Returns a client scoped to the named graph store.

Keyword arguments for chat / stream

AsyncClient mirrors Client with async def methods and async for streaming.

Links

• Daimon on GitHub
• Bug reports

License

Apache-2.0