Conversation traces, not call logs
1chat groups related LLM calls into a single multi-turn trace so product, support, and engineering can inspect the conversation that actually happened.
Unified LLM API for production teams
Outsource your evals.
1chat is the drop-in model gateway that watches real conversations, turns feedback into eval sets, and swaps models when your own data says quality will hold.
OpenAI Chat CompletionsOpenAI ResponsesAnthropic MessagesGemini GenerateContent
Cost + 10%Transparent prepaid margin
4 SDK surfacesOpenAI, Anthropic, Gemini
1 tracePer multi-call conversation
5 routing modesFrom observe to aggressive
Why 1chat exists
Your tax AI company should not need a model-eval team before it gets customers.
If your domain is tax prep, insurance claims, recruiting, tutoring, or clinical admin, the best model is not abstractly “best.” It is the model that preserves quality for your users, your prompt patterns, your tools, your edge cases, and your budget.
Labels become eval sets
Thumbs up, thumbs down, reasons, and inferred follow-up signals become regression datasets without asking your team to build an eval platform first.
Model choice earns trust slowly
Observe, shadow, conservative, balanced, and aggressive modes let cheaper models prove themselves against your data before traffic moves.
Prepaid by design
Every request reserves prepaid balance before provider execution. Customers can top up manually or use auto recharge, but never rack up a surprise bill.
How it works
Start as a compatible gateway. Become a domain-specific routing brain.
The public API remains familiar, while the control plane stores traces, labels, costs, routing decisions, and candidate-model outcomes. Over time, 1chat learns which cheaper or faster models are good enough for each class of task.
Model routing details- Point an existing SDK at 1chat and keep your request shape.
- Add prepaid balance, create project keys, and ship production traffic.
- Review assembled conversation traces and lightweight labels.
- Let 1chat build eval sets and shadow candidate models.
- Move traffic only when your quality floor and aggressiveness policy allow it.
Why 1chat
A model-choice service with observability, evals, and prepaid controls built in.
A compatible gateway that keeps existing SDK request shapes intact.
Conversation-level traces that group multi-call workflows into one inspectable timeline.
Labels and inferred follow-up signals that become customer-specific eval sets.
Routing modes that move traffic only when real production data supports the switch.
Billing
Prepaid balance first. Provider cost plus 10% after completion.
1chat reserves an estimated customer charge before the provider call, finalizes the exact charge when usage returns, and releases unused reservation. Failed calls do not become open invoices.
Manual top-ups and auto recharge
Request-level balance reservation
Ledgered charges, releases, and top-ups