SteerPlane SDK

The safety layer for the AGI era. Runtime guardrails for autonomous AI agents.

Cost limits · Loop detection · Policy engine · Human approval callbacks · SSE streaming gateway · CLI · Docker · 4 framework integrations

Install

pip install steerplane            # Core SDK
pip install steerplane[cli]       # + CLI tool
pip install steerplane[all]       # + CLI + YAML config + LangChain

Quick Start

Option 1: Gateway Mode (Zero Code Changes)

Point your existing OpenAI client to SteerPlane. Every LLM call is automatically monitored, rate-limited, and cost-tracked — including SSE streaming with mid-stream cost enforcement.

import openai

client = openai.OpenAI(
    base_url="http://localhost:8000/gateway/v1",
    api_key="sk_sp_...",  # SteerPlane API key
    default_headers={"X-LLM-API-Key": "sk-..."}
)

# All calls — including stream=True — are now guarded.
response = client.chat.completions.create(
    model="gpt-4o-mini",
    messages=[{"role": "user", "content": "Hello"}],
    stream=True  # SSE streaming with real-time cost tracking
)

Option 2: Decorator Mode

from steerplane import guard

@guard(
    agent_name="support_bot",
    max_cost_usd=10.00,
    max_steps=50,
    denied_actions=["DROP TABLE*", "rm -rf *"],
    rate_limits=[{"pattern": "send_email", "max_count": 5, "window_seconds": 60}],
    require_approval=["execute_trade*"]
)
def run_agent():
    agent.run()

Option 3: Context Manager

from steerplane import SteerPlane

sp = SteerPlane(agent_id="my_bot")

with sp.run(max_cost_usd=10.0, max_steps=50) as run:
    run.log_step("query_db", tokens=380, cost=0.002, latency_ms=45)
    run.log_step("generate_response", tokens=1240, cost=0.008, latency_ms=320)

What's New in v1.1.1

• Built-in pricing now covers 83 models, checked against the OpenAI, Anthropic and Google pricing pages on 21 Oct 2026 (current GPT, Claude and Gemini models, by their API ids).
• Dated model names such as gpt-4o-2024-08-06 are priced as their model instead of the default rate, and gpt-4.1 is no longer priced as gpt-4.
• For a model that is not in the table, pass your own cost per step: run.log_step(..., cost=0.0042).

What's New in v1.1.0

• detect_loops now works. The docs have shown @guard(detect_loops=True) for a long time, but the parameter did not exist. It is accepted by @guard, SteerPlane.run(), RunManager and all four framework integrations, and in .steerplane.yml. Set it to False to stop loop detection from ending a run; cost, step, runtime and policy limits still apply.
• Self-hosting for several teams: see docs/hosted-deployment.md in the repository.

What's New in v0.4.0

🌊 SSE Streaming Gateway

The AI Gateway now supports Server-Sent Events (SSE) streaming. It accumulates token costs per chunk in real time and can terminate a stream mid-response by injecting a steerplane_enforcement event when the cost ceiling is breached — without corrupting the SSE protocol.

🖥️ CLI Tool

steerplane status              # Check API health
steerplane runs list           # List recent runs
steerplane runs inspect <id>   # Inspect a run
steerplane runs kill <id>      # Mark a run as terminated
steerplane keys list           # List API keys
steerplane keys create         # Create new gateway key
steerplane logs --tail         # Live log stream

🐳 Docker Compose

Full 3-service stack in one command (for one server shared by several teams, use docker-compose.hosted.yml):

docker compose up -d
# API (8000) + Dashboard (3000) + PostgreSQL (5432)

📄 Config File (.steerplane.yml)

Project-level defaults auto-discovered from your working directory:

api_url: http://localhost:8000
agent_name: my_bot
defaults:
  max_cost_usd: 10.0
  max_steps: 100
  detect_loops: true

🔌 Framework Integrations

Native support for 4 frameworks — all use lazy imports, so framework dependencies are only required when used:

LangChain:

from steerplane.integrations import SteerPlaneLangChainHandler
handler = SteerPlaneLangChainHandler(agent_name="lc_bot", max_cost_usd=5.0)
llm = ChatOpenAI(model="gpt-4o-mini", callbacks=[handler])

OpenAI Agents SDK:

from steerplane.integrations import SteerPlaneAgentHooks
hooks = SteerPlaneAgentHooks(agent_name="openai_agent", max_cost_usd=10.0)

CrewAI:

from steerplane.integrations import SteerPlaneCrewMonitor
monitor = SteerPlaneCrewMonitor(agent_name="crew_bot", max_cost_usd=15.0)

AutoGen:

from steerplane.integrations import SteerPlaneAutoGenMonitor
monitor = SteerPlaneAutoGenMonitor(agent_name="autogen_bot", max_cost_usd=10.0)

All Features

Architecture

┌─────────────┐     ┌──────────────┐     ┌────────────┐     ┌───────────────┐
│  AI Agent   │────▶│ SteerPlane   │────▶│  FastAPI    │────▶│  PostgreSQL   │
│  (Your App) │     │  SDK / GW    │     │  Server     │     │  Database     │
└─────────────┘     └──────────────┘     └────────────┘     └───────────────┘
                                                │
                                                ▼
                                         ┌───────────────┐
                                         │   Next.js     │
                                         │   Dashboard   │
                                         └───────────────┘

Tech Stack

Links

• GitHub: github.com/vijaym2k6/SteerPlane
• npm: npmjs.com/package/steerplane
• Dashboard: http://localhost:3000
• API Docs: http://localhost:8000/docs

License

MIT — Patent Pending