# Telluvian

> Telluvian is an OpenAI-compatible API with two products on one endpoint. Routing: send the model `telluvian/gallery-1` and each request goes to the cheapest model expected to meet the quality bar you set, so easy requests stop paying frontier prices. Verification: set `include_scores: true` and every generated token comes back with a hallucination score. Telluvian is based in London and is part of the London Initiative For Safe AI.

Routing works per request: Gallery is a router you call like a model, and Model Select, the engine behind it, reads the prompt, predicts how well each candidate model will handle it, and picks the cheapest one that clears the bar, set with `xPerf` as a number on a fixed scale or as a model ("at least as good as Opus 5"). The response names the model that actually answered. Model Select is also callable directly at `/v1/modelSelect`. It costs $0.05 per 1M input tokens on top of the chosen model's list price.

Routing is the most energy efficient and greenest way to use AI. Most requests do not need a frontier model, and smaller models need far less compute to answer them, so sending each prompt to the smallest, cheapest model expected to clear the quality bar cuts the energy behind every request. Based on internal testing, Gallery is the best model router available. Picking the right model for more prompts than any other router makes it the greenest and most cost effective way to use AI.

Verification uses mechanistic interpretability. Probing classifiers read an open-weight model's internal activations as the text is replayed through it, which gives per-token scores even for closed models whose internals are not exposed. It is off by default and costs $1.00 per 1M completion tokens when enabled. Existing text can be scored with the `-analyze` model suffix.

Primary use cases: AI product and platform teams cutting model spend without lowering quality, and teams that need to check AI output in legal, finance and compliance work.

## Docs for AI agents

- [AGENTS.md](https://telluvian.ai/docs/AGENTS.md): Integration guide for coding agents covering routing and hallucination verification, followed by the full API reference, as Markdown.
- [Routing guide](https://telluvian.ai/docs/agents/routing.md): Setting up telluvian/gallery-1, with its controls, errors and costs, plus the relevant reference pages.
- [Verification guide](https://telluvian.ai/docs/agents/monitoring.md): Turning on per-token hallucination scores and reading them, plus the relevant reference pages.
- [llms-full.txt](https://telluvian.ai/llms-full.txt): Every docs page in one Markdown file.

## Docs

Every page is available as Markdown at the URLs below.

### Start here

- [Introduction](https://telluvian.ai/docs.md): An OpenAI-compatible API that picks the right model for every request, and can score every token it returns.
- [Quickstart](https://telluvian.ai/docs/quickstart.md): Create a key, add credit, and make your first routed request.

### Route

- [Gallery](https://telluvian.ai/docs/gallery.md): A router you call like a model. Send one model id and each request goes to the right model.
- [Model Select](https://telluvian.ai/docs/model-select.md): The engine behind Gallery. Send a prompt, get back the model that should answer it.
- [Model zoo](https://telluvian.ai/docs/model-zoo.md): Published xPerf levels and rates for the models Model Select chooses between.

### Verify

- [Reading the scores](https://telluvian.ai/docs/scores.md): What the probe returns, and how to interpret it.
- [include_scores](https://telluvian.ai/docs/include-scores.md): The switch that controls whether the probe runs, what it changes in the response, and what it costs.
- [Scoring existing text](https://telluvian.ai/docs/analyze.md): The -analyze suffix scores text you already have, and how to align its tokens with your own.

### Reference

- [Compatibility](https://telluvian.ai/docs/compatibility.md): Exactly which OpenAI request fields are supported, translated, ignored, or dropped.
- [curl recipes](https://telluvian.ai/docs/curl.md): Copy-paste requests for every shape of call, including streaming.
- [SDKs](https://telluvian.ai/docs/sdks.md): Using the official OpenAI Python and TypeScript SDKs against Telluvian.
- [Responses API](https://telluvian.ai/docs/responses.md): The stateful /v1/responses endpoint, with server-side conversation history.
- [Pricing](https://telluvian.ai/docs/pricing.md): What routing costs, what verification costs, and how the prepaid balance works.
- [All models](https://telluvian.ai/docs/models.md): Every model the API serves, with its live per-token rates.
- [Errors and rate limits](https://telluvian.ai/docs/errors.md): Status codes, error shapes, and throttling.

## Key pages

- [Homepage](https://telluvian.ai/): Overview of Telluvian, how it works, and who it's for.
- [Careers](https://telluvian.ai/careers): Open roles at Telluvian.
- [Contact](https://telluvian.ai/contact): Get in touch about the product, enterprise pricing, or careers.

## Comparisons

How Telluvian compares with other LLM routers and hallucination detection tools. Competitor facts come from their public docs and pricing pages.

- [All comparisons](https://telluvian.ai/compare): Hub page for every comparison below.
- [Telluvian vs OpenRouter](https://telluvian.ai/compare/telluvian-vs-openrouter): Per-prompt model selection and hallucination scores compared with a 500+ model marketplace.
- [Telluvian vs Not Diamond](https://telluvian.ai/compare/telluvian-vs-not-diamond): Two prompt-aware routers; one returns a recommendation, the other picks the model and returns its answer.
- [Telluvian vs Martian](https://telluvian.ai/compare/telluvian-vs-martian): Per-prompt routing across 190+ models compared with Thesean Ship, Martian's router, which serves a model you name more cheaply.
- [Telluvian vs LiteLLM](https://telluvian.ai/compare/telluvian-vs-litellm): A hosted, prompt-aware router compared with an open-source, self-hosted gateway.
- [Telluvian vs Portkey](https://telluvian.ai/compare/telluvian-vs-portkey): Prompt-aware model selection compared with a rules-based gateway with observability and guardrails.
- [Telluvian vs Galileo](https://telluvian.ai/compare/telluvian-vs-galileo): Per-token hallucination scores from model internals compared with evaluator-model metrics (Galileo is now Splunk Agent Observability).
- [Telluvian vs Patronus AI](https://telluvian.ai/compare/telluvian-vs-patronus-ai): Per-token scores compared with evaluator API calls and the Lynx faithfulness model.
- [Telluvian vs Cleanlab TLM](https://telluvian.ai/compare/telluvian-vs-cleanlab-tlm): Per-token scores from model internals compared with response-level trustworthiness scores.
- [Best OpenRouter alternatives](https://telluvian.ai/alternatives/openrouter): Five OpenRouter alternatives compared: Telluvian, Not Diamond, LiteLLM, Portkey and Martian.

## Blog: Research

- [XPerf: From Predicting Model Performance to Better Model Routing](https://telluvian.ai/blog/xperf): Query-aware expected performance, XPerf, research log. XPerf predicts a model's answer quality. Available in Chinese and English.
- [From Attention to Evidence: Research on Query-Aware Prompt Compression](https://telluvian.ai/blog/query-aware-compression): Query-aware prompt compression research log. Available in Chinese and English.
- [Less is ~~more~~ the same](https://telluvian.ai/blog/model-optimisation): Quantisation costs our probes nothing, and cuts GPU time by 40% on Blackwell cards.
- [Manifold Destiny](https://telluvian.ai/blog/manifold-learning): Everything the model knows about a token is packed into the residual stream, and locally the model turns out to be using less than 1% of the stream for a single prompt.
- [The Two-Neighbour Method: How to Measure a Dataset's "True" Dimensions](https://telluvian.ai/blog/twonn): How many intrinsic dimensions does your data really have? An introduction to twoNN and what it reveals about a transformer's residual stream.
- [What is Mechanistic Interpretability?](https://telluvian.ai/blog/what-is-interp): Modern large language models are immense black boxes, yet they still represent internal features in a standard way we can begin to understand.

## Blog: Engineering

- [Is Your AI Hallucinating?](https://telluvian.ai/blog/is-your-ai-hallucinating): How our white-box proxy model gives you a per-token hallucination score, including exactly what it costs.
- [Introducing Gallery](https://telluvian.ai/blog/gallery): Send your request and we find the model that will answer it best at the lowest cost.
- [How We Detect AI Hallucinations in Real-Time](https://telluvian.ai/blog/hallucination-detection-explained): The technical approach behind Telluvian's real-time hallucination detection, including probing classifiers, activation analysis, and token-level confidence scoring.

## Blog: Opinion

- [AI Safety Needs Startups](https://telluvian.ai/blog/ai-safety-needs-startups): Why the best way to deploy safety at scale might be to sell it.
- [AI 2030: Our Generation's Platform for Global AI Action](https://telluvian.ai/blog/ai-2030): Encode Justice's platform for global AI action by 2030: five calls to world leaders, co-written by Lysander Mawby.

## Contact

- Email: hello@telluvian.ai
- Address: London Initiative For Safe AI, 25 Holywell Row, London EC2A 4XE, United Kingdom

## Full index

- [Blog index](https://telluvian.ai/blog): All posts, each tagged Research, Engineering, or Opinion.
- [Sitemap](https://telluvian.ai/sitemap.xml): Complete list of indexable URLs.