How do I use a custom OpenAI-compatible model in Pydantic AI?

Construct an OpenAIChatModel with an OpenAIProvider whose base_url is https://codeburst.ai/api/v1 and api_key is your CodeBurst key, then pass it to Agent. The model name is codeburst-agent.

Why does a typed Pydantic AI agent need failover?

Structured-output and tool-using agents depend on every turn returning valid content. An empty tool-synthesis turn or a provider rate limit breaks the run before the typed result is produced. CodeBurst fails over across providers and repairs empty tool turns so the agent completes its contract.

Does this change my Pydantic models or tools?

No. Only the model/provider config changes; your result_type, tools and validation stay the same.

Guide · Updated June 2026

The best LLM for Pydantic AI

Pydantic AI is for people who want their agents typed — validated outputs, structured tool calls, production discipline. But strict structure has a sharp edge: if a turn returns empty, or a provider rate-limits mid-run, the typed contract breaks before you get a result. CodeBurst gives Pydantic AI an OpenAI-compatible model that fails over across providers and repairs empty tool turns, so your typed pipeline actually returns its type.

Structure makes reliability non-negotiable

A loose chatbot can shrug off a weird response. A typed agent can't — it expects a valid tool call or a parseable structured output every turn. Two failures hit harder here:

Empty tool-synthesis turns. A reasoning model returning blank on a tool turn doesn't just look odd — it fails validation and aborts the run.
Rate limits mid-run. A provider cap between turns kills the request before the structured result is assembled.

What CodeBurst adds

Failure	CodeBurst
Provider rate-limits mid-run	Reroutes to a healthy provider in the same request; the turn returns valid content.
Empty tool-synthesis turn	`codeburst-agent` retries with a corrected format, preserving the typed contract.
A critical step needs more rigor	Use `codeburst-swarm` for a multi-model vote on that step.

Configure the model

Point an OpenAIProvider at CodeBurst:

from pydantic_ai import Agent
from pydantic_ai.models.openai import OpenAIChatModel
from pydantic_ai.providers.openai import OpenAIProvider

model = OpenAIChatModel(
    "codeburst-agent",
    provider=OpenAIProvider(
        base_url="https://codeburst.ai/api/v1",
        api_key="YOUR_CODEBURST_KEY",
    ),
)

agent = Agent(model, result_type=YourSchema)  # your tools + types, unchanged

Your result_type, tools and validation don't change — the endpoint just gains provider failover and tool-call repair underneath.

Get started

Get an API key The best tool-calling API →

FAQ

How do I use a custom model in Pydantic AI?
OpenAIChatModel("codeburst-agent", provider=OpenAIProvider(base_url="https://codeburst.ai/api/v1", api_key=...)).

Why does a typed agent need failover?
Empty turns and rate limits break the typed contract; CodeBurst keeps every turn returning valid content.

Does my schema change?
No — only the model/provider config.