Why does a CrewAI crew need failover?

A crew runs several agents, each making model calls, so total call volume is high and bursty — exactly what trips provider rate limits. CodeBurst fails over across providers and repairs empty tool turns so the crew completes its work.

Do I set the LLM per agent or per crew?

Either. Pass the LLM instance to individual agents or set it as the crew default; both point at the same CodeBurst endpoint.

Guide · Updated June 2026

The best LLM for CrewAI

Q: How do I set a custom LLM in CrewAI?

Use CrewAI's LLM class with model set to openai/codeburst-agent, base_url to https://codeburst.ai/api/v1, and your CodeBurst key. CrewAI uses LiteLLM for custom providers, so the openai/ prefix routes to the OpenAI-compatible endpoint.

A CrewAI crew is several agents working a task together — a researcher, a writer, a reviewer — each making its own model calls. That parallelism is the point, but it also multiplies your call volume, and bursty volume is exactly what trips a provider's rate limit. CodeBurst gives the whole crew an OpenAI-compatible LLM that fails over across providers and repairs empty tool turns, so one busy provider doesn't stall the crew.

More agents, more calls, more failure surface

Scale a crew from one agent to five and you've roughly five-x'd your calls per task — often in tight bursts as agents hand off. Two things end a crew run early:

Rate limits under burst load. Several agents calling at once trip a per-minute cap, and an agent's step fails.
Empty tool turns. A reasoning model returns blank on a tool-synthesis turn and that agent stalls, blocking the crew.

What CodeBurst adds

Failure	CodeBurst
Provider rate-limits under burst	Reroutes to a healthy provider in the same request; the agent's step succeeds.
Empty tool-synthesis turn	`codeburst-agent` retries with a corrected format.
One agent needs a stronger brain	Give that agent `codeburst-swarm` for a multi-model vote.

Configure the crew's LLM

CrewAI uses LiteLLM for custom providers, so prefix the model with openai/:

from crewai import Agent, LLM

llm = LLM(
    model="openai/codeburst-agent",
    base_url="https://codeburst.ai/api/v1",
    api_key="YOUR_CODEBURST_KEY",
)

researcher = Agent(role="Researcher", goal="...", llm=llm)
# pass `llm` to each agent, or set it as the crew default.

Your agents, tasks and tools are unchanged — only the LLM endpoint moves, and the crew gains provider failover and tool-call repair.

Get started

Get an API key How agent failover works →

FAQ

How do I set a custom LLM in CrewAI?
Use LLM(model="openai/codeburst-agent", base_url="https://codeburst.ai/api/v1", api_key=...) and pass it to your agents.

Why does a crew need failover?
Multiple agents = high bursty call volume that trips rate limits; CodeBurst fails over and repairs tool turns.

Per agent or per crew?
Either — same CodeBurst endpoint.