Why CodeBurst

A router that heals itself.

Most gateways just proxy your request to one provider. When that provider rate-limits or goes down, you get the error. CodeBurst is built around the assumption that providers fail — and routes around it for you.

🔁

Automatic failover

Every model is a fallback chain, not a single endpoint. The instant a provider rate-limits or degrades, CodeBurst reroutes mid-request to the next healthy lane. Nightly health probes keep the chains fresh, so a dead provider is dropped before it ever reaches you.

🎯

Task-aware routing

Don't pick a model — pick a job. codeburst-agent for tool-calling, codeburst-vision for images, codeburst-compress for context, codeburst-swarm for hard reasoning. Each alias routes to the model that's measurably best at that task.

🔌

Drop-in OpenAI API

Standard /v1/chat/completions. Change your base_url and key — keep your existing OpenAI SDK, your code, your tooling. Swap models by changing one string, never your integration.

💸

No markup

Bring your own provider keys and pay the provider directly — CodeBurst adds zero per-token markup. Or hand us the keys and let managed routing pick the best model per request. Your bill, your control.

🧩

Provider breadth

NVIDIA NIM, Mistral, Google Gemini, Groq, Cohere, SambaNova, Cerebras, Cloudflare and more — behind one endpoint. New providers slot into the routing layer without a single change to your app.

🛡️

Reliability patterns built in

Beyond failover: multi-model voting, debate, and size-aware context routing ship as first-class aliases. Reach for them with a model name — no orchestration code on your side.

Drop-in

Two lines to switch. Zero to your code.

If your app already speaks the OpenAI API, you're done in under a minute.

python

from openai import OpenAI

client = OpenAI(
    base_url="https://codeburst.ai/api/v1",   # ← the only change
    api_key="YOUR_CODEBURST_KEY",
)

resp = client.chat.completions.create(
    model="codeburst-best",      # or codeburst-agent / codeburst-vision / codeburst-swarm
    messages=[{"role": "user", "content": "Hello"}],
)
print(resp.choices[0].message.content)

Coming from OpenRouter?

Same pattern. More resilience.

CodeBurst speaks the exact same OpenAI-compatible interface — so migrating is a base_url change. What you gain is routing that actively works around provider failures instead of passing them through.

CodeBurst

Every model is a multi-provider fallback chain
Mid-request reroute when a provider degrades
Nightly health probes prune dead lanes automatically
Task-aware aliases (agent · vision · compress · swarm)
Built-in voting / debate / size-aware context routing
BYOK with no markup, or managed routing

Typical router

Routes to one provider per request
Provider error is returned to your app
Manual model selection per call
You build orchestration / retries yourself
Markup on managed credits
Switching models is on you

Models

Popular models, one endpoint.

Call any model by name, or use a smart alias and let CodeBurst route to the best one for the job — with failover already wired in.

Nemotron Super 120B Qwen3.5 397B DeepSeek V4 GPT-OSS 120B Mistral Large Mistral Small Llama 3.3 70B Llama 4 Scout Gemini 2.5 Flash Gemini 2.5 Pro Command R Kimi K2 GLM 5.1

Smart aliases — route by job, not by model

Pick the task; CodeBurst picks (and fails over between) the models that are measurably best at it.

codeburst-bestTop-quality reasoning — the strongest model currently up

codeburst-agentTool-calling & multi-step agents, with tool-format repair built in

codeburst-visionImages — receipts, screenshots, photos (gpt-4o → gemini → scout)

codeburst-swarmHard problems — multiple models vote for a more reliable answer

codeburst-compressSize-aware context compaction — never overflows a model's window

codeburst-fastLow-latency lane for quick, cheap turns

codeburst-1m / 2mMillion-token+ context for long documents

codeburst-128k / 256kPick a context tier explicitly

One API for every model.
It routes itself.