Why does a smolagents run stall?

A code agent loops — write code, run it, read the result, iterate — making many model calls. Any one can hit a provider rate limit or return an empty tool turn. CodeBurst fails over across providers and repairs empty tool turns so the loop finishes.

Does CodeBurst work with smolagents tool calling?

Yes. codeburst-agent handles the OpenAI tools format and includes tool-call repair, so the agent's tool steps don't return empty and stall.

Guide · Updated June 2026

The best LLM for smolagents

Q: How do I use a custom model with smolagents?

Use OpenAIServerModel with model_id set to codeburst-agent, api_base set to https://codeburst.ai/api/v1, and your CodeBurst key. It calls any OpenAI-compatible endpoint.

smolagents takes a simple idea — let the agent think in code, run it, and iterate — and that loop is where reliability shows. A CodeAgent writes a snippet, executes it, reads the result, and goes again, often many times per task. One provider rate-limit or empty tool turn mid-loop and the run halts. CodeBurst gives smolagents an OpenAI-compatible model that fails over across providers and repairs empty tool turns, so the agent keeps iterating to a solution.

Code-and-iterate means many calls

The whole point of a code agent is the loop: act, observe, refine. Each pass is a model call, so even a "small" agent can make a lot of them on a real task — and call count is what drives the odds of hitting a provider's rate limit or an empty tool-synthesis turn that quietly stalls the run.

What CodeBurst adds

Failure	CodeBurst
Provider rate-limits mid-loop	Reroutes to a healthy provider in the same request; the step completes.
Empty tool-synthesis turn	`codeburst-agent` retries with a corrected format.
A hard step needs more reliability	Point that model at `codeburst-swarm` for a multi-model vote.

Configure the model

smolagents calls any OpenAI-compatible endpoint via OpenAIServerModel:

import os
from smolagents import OpenAIServerModel, CodeAgent

model = OpenAIServerModel(
    model_id="codeburst-agent",
    api_base="https://codeburst.ai/api/v1",
    api_key=os.environ["CODEBURST_API_KEY"],
)

agent = CodeAgent(tools=[...], model=model)
agent.run("...")

Your tools and agent logic stay the same — the model endpoint just stops depending on one provider being up for the whole loop.

Get started

Get an API key Best LLM for AI agents →

FAQ

How do I use a custom model with smolagents?
OpenAIServerModel(model_id="codeburst-agent", api_base="https://codeburst.ai/api/v1", api_key=...).

Why does a run stall?
The code loop makes many calls; CodeBurst fails over and repairs empty tool turns so it finishes.

Does tool calling work?
Yes — codeburst-agent handles tools with tool-call repair.