Tracer is an AI research lab building coordinated AI systems from ensembles of specialist open-source models that work together, helping AI teams improve performance per token through intelligence allocation and coordination.
Tracer is an AI research lab building more capable and efficient AI through model coordination. We research adaptive inference, ensembling, and systems in which multiple open-source models work together to produce capabilities beyond any single model. Our work spans reasoning, code, research, and agentic systems, with the goal of making advanced intelligence more accessible and economical.
Open-source classification-routing project; trains a lightweight surrogate from model production traces to serve predictable inputs locally and defer uncertain cases to an LLM t...
What customers say
Tracer reports that Obside deployed a 38-cell surrogate behind an OpenAI-compatible endpoint to match news headlines against user intents, with fallback models for uncertain inputs. The vendor reports 95% savings versus…
Tracer reports that getclaw uses a native Hermes plugin to route tool-selection decisions through a local TRACER classifier, deferring uncertain cases to an LLM. On measured production traces, end-to-end agent cost fell…
Tracer Launches Observability Decorator and Cloud CLI in v0.3.0
Tracer has released version 0.3.0, introducing the 'tracer.watch' decorator. This tool observes existing LLM pipeline calls and ships them to Tracer Cloud for optimization. The update includes a corresponding zero-dependency JavaScript/TypeScript mirror for browser-based observability, alongside a new command-line interface for managing tracers, training, routing, and analytics within the Tracer Cloud ecosystem. The release also improves the robustness of the 'tracer scan' functionality during data uploads.
New feature
Tracer Releases Version 0.2.0 with Scan and Safety Gate Features
Tracer version 0.2.0 introduces 'tracer scan,' a tool that analyzes existing LLM production traces to estimate potential cost savings and coverage before training begins. The release also upgrades the platform's routing capabilities with a new parity certification mechanism that uses held-out data sets to ensure reliable model performance, and an out-of-distribution (OOD) safety gate that redirects anomalous inputs to a teacher model. These updates aim to improve the accuracy and safety of automated cost-efficient LLM routing.
