Serving agents

Expose any agent over HTTP. Requires the distributed extra:

pip install "maki-framework[distributed]"

Serve

# agent.yaml
name: MyAgent
model: llama3.2
role: assistant
# backend: ollama | llamacpp | openai | anthropic | openrouter (default: ollama)
# base_url: http://127.0.0.1:8080/v1   # llamacpp only
plugins:
  - web_search
  - file_reader

maki serve --config agent.yaml --host 127.0.0.1 --port 8100

Options: --api-key KEY (bearer-token auth), --tls-cert FILE and --tls-key FILE (HTTPS). The server exposes /execute, /stream, /health and memory endpoints.

Connect from another process

from maki.distributed.proxy import AgentProxy

agent = AgentProxy(name="MyAgent", base_url="http://127.0.0.1:8100")
print(agent.execute_task("Summarize the latest AI news."))

AgentProxy has the same surface as Agent and includes a per-proxy circuit breaker.

Mix local and remote agents

DistributedAgentManager lets one workflow use both:

from maki.distributed.registry import DistributedAgentManager

Workflow checkpoints can be stored as local JSON files (LocalStateStore) or in Redis (RedisStateStore, extra distributed-redis).

Warning

Bind to 127.0.0.1 unless you also set --api-key and TLS. An agent with plugins can read files and reach the network on behalf of whoever can call it.