# Tracer > Tracer is a Y Combinator backed research lab helping AI teams improve > performance per token through intelligence allocation and coordination. > Echo is Tracer's latest product, available through one OpenAI-compatible > endpoint for chat, reasoning, code, and agent harnesses. Canonical site: https://tracerml.ai/ Last updated: 2026-08-26 ## Product - [Echo](https://echo.tracerml.ai/): Use Echo in the hosted chat interface. - [Echo API documentation](https://echo.tracerml.ai/docs/api): Connect through the OpenAI-compatible API, including from OpenCode and compatible agent harnesses. - [Echo API keys](https://echo.tracerml.ai/account#api-keys): Create and manage API access. Echo presents one public model named `echo`. Behind that stable interface, the system allocates different capabilities and amounts of computation according to the work. Users do not choose internal models or modes. ## Research thesis - Intelligence is a system property. - Organization can create capability. - An ensemble matters only when the coordinated whole beats its strongest measured component. - Quality, latency, and cost must be evaluated end to end in the same harness. - Self-organizing intelligence is a research direction, not a current product claim. Tracer studies coordinated intelligence, adaptive inference, interpretable orchestration, and agentic systems. The work targets more independent, useful evidence per dollar. ## Work with Tracer Tracer works with teams that care about inference cost, token efficiency, and model performance. The lab applies intelligence allocation and coordination to help teams get better outcomes with less wasted computation. - [Book a 30-minute inference review](https://cal.com/adam-rida-tracer/30min) ## Research lineage - [Pricing Capability](https://tracerml.ai/research/pricing-capability/): A research article examining Google's selected $10 million bid for a Spirit Aviation package of deidentified operational records and internally developed software. Capability Alpha measures the task improvement attributable to data under controlled evaluation. Price-Implied Uplift works backward from an observed price. Under the paper's assumptions, the Spirit case implies a 0.236% threshold. Three recurring public-company agreements form a standardized 0.58%-1.42% reference range; conditional buyer-specific scenarios span approximately $50 million to $200 million. - [Pricing Capability paper](https://tracerml.ai/research/pricing-capability/pricing-capability-paper.pdf): The complete 21-page final submission, dated August 25, 2026. It includes the Spirit valuation scenarios and public-transaction calculations. - [TRACER: Learning When to Defer](https://tracerml.ai/research/tracer-oss/): A plain-language and technical account of trace-based, parity-gated routing for fixed-label LLM classification. - [TRACER open source](https://github.com/adrida/tracer): Trace-driven, parity-gated inference and interpretability tools. - [TRACER paper](https://arxiv.org/abs/2604.14531): Trace-Based Adaptive Cost-Efficient Routing for LLM Classification. - [Research programs](https://tracerml.ai/#research): Current research themes. ## Evidence and company information - [Case studies](https://tracerml.ai/case-studies/): Dated production evidence from Tracer's earlier efficiency work. - [Research notes](https://tracerml.ai/blog/): Technical and product writing. - [Changelog](https://tracerml.ai/changelog/): Dated updates. - [GitHub](https://github.com/adrida/tracer): Public code and documentation. - [Book a call](https://cal.com/adam-rida-tracer/30min): Discuss inference cost, token efficiency, and model performance with the team. ## Naming - Organization: Tracer - Product: Echo - Public API model identifier: `echo` - Primary domain: `tracerml.ai` - Echo domain: `echo.tracerml.ai` Echo exposes one public model and one endpoint. Its internal allocation is not described as a fixed number of models, a simple switch, or autonomous self-organization. TRACER OSS is Tracer's open-source classification-routing project; Echo is Tracer's current adaptive AI system.