|
Download docs/adr/0001-python-deepagents-rewrite.md from build-small-hackathon/agent-swarm-workbench: direct link, hf CLI and curl.
- Browser
- Download file 950 Bytes
-
https://huggingface.co/spaces/build-small-hackathon/agent-swarm-workbench/resolve/main/docs/adr/0001-python-deepagents-rewrite.md
- Command line
-
hf download hf://spaces/build-small-hackathon/agent-swarm-workbench/docs/adr/0001-python-deepagents-rewrite.md
-
curl -L -o 0001-python-deepagents-rewrite.md https://huggingface.co/spaces/build-small-hackathon/agent-swarm-workbench/resolve/main/docs/adr/0001-python-deepagents-rewrite.md
950 Bytes
A newer version of the Gradio SDK is available: 6.30.0
ADR 0001: Rewrite Around Python DeepAgents
Status
Accepted
Context
Previous app mixed SvelteKit, Cloudflare Durable Objects, E2B sandboxes, Vercel AI SDK, OpenRouter, and a browser game UI. New direction is ground-up Python using LangChain DeepAgents and their sandbox backend interface.
Decision
Use Python FastAPI for HTTP, LangChain DeepAgents for contestant-side autonomous building, LangGraph for judging workflows, OpenRouter for model access, and Daytona as default DeepAgents sandbox provider.
Keep sandbox provider behind DEEPAGENT_SANDBOX_PROVIDER so local smoke tests can use LocalShellBackend explicitly.
Consequences
- Legacy TS/Svelte/Worker/E2B modules are removed.
- Main contestant-side seam is
arena.agent.create_session(). - Judging should live behind a separate LangGraph seam.
- Route handlers do not know provider details.
- Local backend is allowed only for smoke tests because it is not isolated.