Documentation v1.0
What is Reliant?
Reliant is an infrastructure layer that sits between your application and any LLM, ensuring structured outputs, intelligent retry, and complete observability in production.
The problem we solve
Any developer who puts LLMs into production faces the same problems: JSON that fails to parse, schemas that fail silently, zero visibility into what is breaking and why.
The common solution is to write manual retry logic in each project — fragile, inconsistent and untested. Reliant solves this once and for all with a simple API.
💥
Without Reliant
Invalid JSON breaks the pipeline. Manual retry in every project. Zero visibility in production.
🛡️
With Reliant
Always valid output. Automatic retry with rewritten prompt. Complete log of every execution.
How it works
Reliant intercepts each LLM call, validates the output against the defined schema, and retries intelligently when needed — all transparent to your application.
1
You define a Schema
A JSON Schema that describes the expected output contract. Reliant ensures the LLM always returns in this format.
2
You call reliant.execute()
Instead of calling the LLM directly, you go through Reliant with your prompt, schema, and chosen provider.
3
Reliant ensures the output
If the LLM returns invalid output, Reliant rewrites the prompt with the errors and retries — up to 3x before using the safe fallback.
4
Everything logged in the Dashboard
Every execution recorded — attempts, latency, tokens, validation errors. Complete visibility in production.
Supported providers
Reliant is provider-agnostic. You can use any combination:
| Provider | Tested models | Status |
|---|---|---|
| anthropic | claude-sonnet-4, claude-opus-4, claude-haiku | ✅ Supported |
| openai | gpt-4o, gpt-4o-mini, gpt-4-turbo | ✅ Supported |
| gemini | gemini-1.5-pro, gemini-1.5-flash | ✅ Supported |