DotCode SDK — Python

Embed the DotCode coding agent in your application and drive it with any LLM (Anthropic, OpenAI, Azure OpenAI, Gemini, DeepSeek, Ollama, OpenAI-compatible). The SDK talks JSON-RPC to a dotcode serve process.

Built by Gravicode Studios, led by Kang Fadhil. · 🇮🇩 Dokumentasi Bahasa Indonesia: docs/id/sdk.md

Install

pip install dotcode-sdk

The SDK needs the dotcode CLI: put it on PATH or set DOTCODE_CLI_PATH (a path to dotcode.dll is started with dotnet). Configure providers with environment variables (ANTHROPIC_API_KEY, OPENAI_API_KEY, DEEPSEEK_API_KEY, …), ~/.dotcode/settings.json, or per session with the settings option.

Usage

import asyncio
from dotcode_sdk import DotCodeClient, tool

@tool("now", "Current time")
def now(args):
    import datetime
    return datetime.datetime.now().isoformat()

async def main():
    async with DotCodeClient() as client:          # spawns `dotcode serve`
        session = await client.create_session(model="deepseek:deepseek-v4-flash", tools=[now],
                                              on_permission_request=lambda r: {"decision": "allow"})
        async for e in session.stream("What time is it? Then list the files here."):
            if e["type"] == "assistant.text.delta":
                print(e["text"], end="")

asyncio.run(main())

Features

• Sessions with streaming events (text deltas, tool calls, diffs, todos, usage, cost)
• Custom tools implemented in your app
• Permission handler — sessions are deny-by-default without one; decisions: allow, allow_always, allow_session, deny (with feedback)
• Question (AskUserQuestion) and plan-review handlers
• MCP servers, allowed/disallowed tool rules, permission modes, system prompt overrides, model switching, compaction, transcripts, resume/fork

Full guide: docs/en/sdk.md · Protocol: schema/protocol.schema.json

License: MIT