New: route GPT-5.6 Sol, Terra, and Luna through Seldon. Choose frontier, balanced, or high-volume routing behind one API.
View models →Combine AI with proven software techniques.
Make production AI faster, more efficient and more reliable.
Seldon starts as a drop-in LLM router. Over time it infers the business process behind your traffic and synthesizes recurring work into cheaper, deterministic execution paths — all behind the same API.
Open beta
Start with the Seldon open beta.
Get early access to the router and compiler. Bring your existing LLM traffic, watch recurring work turn into cheaper pipelines, and help shape v0.
Open Beta
$25 starter credit for anyone. Instant access, no card required.
Startup Program
Up to $500 in credits for startups with real LLM workloads. Application-based after signup.
Create a beta account first. Startup applications are reviewed after signup.
Why Seldon
Too many production LLM calls are just expensive data work. Seldon finds the repeatable ones and helps you evaluate deterministic paths.
Cost management.
Reduce frontier LLM usage by 50–80%, with minimal to no compromise in application performance.
Effortless productionization.
Use detected workflow evidence to decide what is worth evaluating before a real compiled path is built.
Reliability and auditability.
Move from non-deterministic prompt chains to versioned data contracts and explainable pipelines built for mission-critical and regulated use cases.
Observability.
Get clarity on how your LLM workflows are driving your business processes and application logic.
Turn repeated LLM calls into production data pipelines.
See how it works →Drop-in router
Point your base_url at Seldon.
Route LLM calls through an OpenAI-compatible endpoint with cost, latency, and fallback controls. No SDK changes — swap the base URL and keep shipping.
Seldon base URL
SELDON_BASE_URL=https://api.seldon-ai.comUse this with an OpenAI-compatible SDK; it adds the /v1 API path.
Trace intelligence
See the workflow evidence behind the traffic.
Live Audit periodically groups similar routed calls into detected workflows and shows the evidence behind a possible replacement — not just per-request logs.
Program synthesis
Evaluate recurring work as compiled paths.
A possible replacement is an audit recommendation, not an artifact. When a path is actually built, Compiled paths tracks its evaluation, shadow testing, activation, or rollback.
FAQ
Questions from teams with real LLM traffic.
The short version: start with routing, let Live Audit identify repeated workflows, and only promote compiled paths when the evidence is strong enough.
What is Seldon?
Seldon is the compilation layer for production AI: you keep calling one API, and over time repeated LLM work becomes faster, cheaper, and more reliable software. Today you start with a drop-in router and Live Audit; the long-term product compiles those workflows into production data pipelines behind the same API.
How is Seldon different from a normal AI gateway?
Most gateways unify model access, routing, budgets, and logs. Seldon starts there, then aims to learn when a call should not be a model call at all — and replace repeated work with a compiled path (SQL, Python, classical NLP, small models, cache) instead of another frontier LLM.
What's the long-term vision?
Seldon becomes an AI compiler and runtime: Application → Router → Compiler → hybrid execution, with frontier models only when necessary. You still think you are calling an LLM; increasingly the answer comes from an explicit, inspectable execution plan.
Why combine AI with proven software techniques?
LLMs are the fastest way to explore ambiguous workflows. Production needs determinism, auditability, and lineage. Seldon's job is to graduate repeated AI behavior into governable software — so you get the speed of AI with the reliability of ordinary data pipelines.
How do I start using it?
Create a Seldon API key, point your OpenAI-compatible client at the Seldon base URL, and send traffic through the router. Route traffic now; optimization comes from observed patterns over time. Your existing OpenAI SDK, LiteLLM, LangChain, or Vercel AI SDK integration keeps the same client shape.
What is Live Audit?
Live Audit is the evidence step toward compilation. It periodically analyzes routed calls for one API key and period, groups similar calls into detected workflows, and may suggest a cheaper path to evaluate. It shows where expensive data processing is hiding — it does not build or deploy a replacement yet.
How does optimization actually roll out?
Observe traffic, surface repeated workflows, shadow a candidate path, promote only when confidence is high, and fall back to the frontier model when it is not. Open-ended reasoning and creative work stay on the frontier. Early audit savings are projected; measured savings come only after a compiled path serves traffic.
What data does Seldon retain?
Seldon retains API data needed for billing, reliability, debugging, audits, cache visibility, and compiled-path analysis. Trace payload capture redacts sensitive headers and secret-like values before storage.
Do you train models on customer data?
No. Seldon does not use customer prompts, outputs, traces, or cached responses to train generalized AI models.
Can we try Seldon safely in staging first?
Yes. Route staging or a limited production key first, confirm requests in Calls & usage and Live Audit, then expand once the integration and reporting look right.