edgecrab 0.5.0

Python SDK for EdgeCrab — a Rust-native autonomous coding agent

edgecrab-sdk

Python SDK for EdgeCrab — a Rust-native autonomous coding agent.

Install

pip install edgecrab-sdk

Quick Start

from edgecrab import EdgeCrabClient

# Connect to a running EdgeCrab API server
client = EdgeCrabClient(
    base_url="http://127.0.0.1:8642",
    api_key="your-api-key",  # optional
)

# Simple chat
reply = client.chat("Explain Rust ownership in 3 sentences")
print(reply)

# With system prompt
reply = client.chat(
    "Refactor this function",
    system="You are a senior Rust developer",
    model="anthropic/claude-sonnet-4-20250514",
)

Async

import asyncio
from edgecrab import AsyncEdgeCrabClient

async def main():
    async with AsyncEdgeCrabClient() as client:
        reply = await client.chat("Hello!")
        print(reply)

asyncio.run(main())

Agent API (recommended)

from edgecrab import Agent

agent = Agent(
    model="anthropic/claude-sonnet-4-20250514",
    system_prompt="You are a helpful coding assistant",
)

# Chat with automatic conversation history
reply = agent.chat("Explain Rust ownership")
print(reply)

# Continue the conversation
follow_up = agent.chat("Give me an example")
print(follow_up)

# Full run with result metadata
result = agent.run("Refactor this function")
print(result.response)
print(f"Turns: {result.turns_used}, Tokens: {result.usage.total_tokens}")

Streaming

from edgecrab import EdgeCrabClient, ChatMessage

with EdgeCrabClient() as client:
    messages = [ChatMessage(role="user", content="Write a haiku about Rust")]
    for chunk in client.stream_completion(messages=messages):
        for choice in chunk.choices:
            if choice.delta.content:
                print(choice.delta.content, end="", flush=True)
    print()

CLI

edgecrab chat "What is the meaning of life?"
edgecrab chat --model gpt-4 --system "Be concise" "Explain monads"
edgecrab chat --stream "Tell me a story"
edgecrab models
edgecrab health

Environment Variables

Variable	Description
EDGECRAB_BASE_URL	API server URL (default: http://127.0.0.1:8642)
EDGECRAB_API_KEY	Bearer token for authentication

API Reference

EdgeCrabClient

Method	Description
chat(message, *, model, system, temperature, max_tokens)	Simple chat — returns string
create_completion(messages, *, model, temperature, max_tokens, tools)	Full completion — returns ChatCompletionResponse
stream_completion(messages, *, model, temperature, max_tokens, tools)	Streaming — yields StreamChunk
list_models()	List available models
health()	Health check

AsyncEdgeCrabClient

Same API as EdgeCrabClient, but all methods are async.

Agent / AsyncAgent

Method	Description
chat(message)	Send message, return reply. Maintains history.
run(message, *, max_turns)	Full conversation run — returns AgentResult
add_message(role, content)	Inject a message into history
reset()	Clear history, start new session
get_messages()	Get conversation history
get_turn_count()	Number of completed turns
get_usage()	Accumulated token usage
list_models()	List available models
health()	Check server health

Links

• GitHub
• Node.js SDK
• EdgeCrab Documentation