aice-agent-platform

aice-agent-platform

Python SDK for the Agent Platform the config-driven LangGraph agent runtime exposed over Aegra's Agent Protocol.

This SDK is a composition of two surfaces:

1. AICE-specific management routes (agents, skills, usage, suggestions) go through aice-core's shared transport, so they get the usual typed AiceError hierarchy, retries, and idempotency-key handling.
2. The native LangGraph surface (threads, runs, assistants, crons) this service is a LangGraph Platform-compatible deployment, so this SDK composes the official langgraph-sdk directly rather than re-implementing it. This is where streaming/run execution lives.

These two surfaces do not share error types or retry behavior. client.threads/client.runs/client.assistants/client.crons raise whatever langgraph-sdk itself raises, with its own retry policy never aice_core.AiceError. client.agents/client.skills/client.usage/ client.suggestions behave like every other AICE SDK. See the docstring at the top of client.py for the full rationale.

from aice_agent_platform import AgentPlatformClient

client = AgentPlatformClient(base_url="https://agents.example.com", api_key="sk-…")

# ── AICE-specific management routes ────────────────────────────────────

# List available agents
agents = client.agents.list()

# Publish a saved agent definition so it's runnable
client.agents.publish("ag_123")

# Discover skills
skills = client.skills.list()

# Fetch usage counters for the current project
usage = client.usage.get()

# Trigger the post-turn suggestion engine for a thread
suggestions = client.suggestions.create_for_thread("th_123")

# ── Native LangGraph surface (streaming, threads, runs) ────────────────

thread = client.threads.create()

for chunk in client.runs.stream(
    thread["thread_id"],
    assistant_id="agent",
    input={"messages": [{"role": "user", "content": "hi"}]},
):
    print(chunk.event, chunk.data)

Async variant mirrors this with AsyncAgentPlatformClient and await/ async for.

See the upstream service's own docs/API.md for the AICE-specific endpoint spec this SDK tracks, and the LangGraph SDK reference for the native threads/runs/assistants/crons surface.