Log10 (log10.io) is an AI accuracy platform founded in 2023 by Arjun Nair (formerly of Nervana/Intel) and Niklas Nielsen (formerly Head of Product at MosaicML). It sits between foundational model providers (OpenAI, Anthropic, Gemini, Mistral) and the applications that call them, capturing every LLM call for logging, evaluation, and continuous improvement. The core product centers on three capabilities: structured logging of LLM calls across multiple providers with a single-line, proxy-less integration; an evaluation and human review workflow (an 'Inbox') where annotations from SMEs and end-user feedback flow into an automatic dataset-curation pipeline; and AutoFeedback, a system that generalizes from a small set of expert-annotated samples to score completions at scale with expert-level precision, providing a real-time accuracy signal. Log10 is designed for engineering and product teams in organizations deploying LLMs in production, especially those in regulated or high-stakes industries where output quality must be measurable and auditable. It supports agentic frameworks including LangChain and Magentic, and integrates with providers including OpenAI, Anthropic, Gemini, Mistral, Together, Lamini, and MosaicML. The platform uses AutoFeedback output to drive downstream model improvement via techniques such as DPO (Direct Preference Optimization) and RLHF, closing a feedback loop from production logs to fine-tuned models. The company raised a $7.2M seed round led by TQ Ventures and Quiet Capital, with hundreds of organizations reported as users including Echo AI, OpenBB, Lime, and Doppler. Key features: - Proxy-less, one-line integration with OpenAI, Anthropic, Gemini, Mistral, Together, and other providers - AutoFeedback: scales expert-level LLM completion evaluation from a small set of annotated samples - Human review Inbox for annotating completions and collecting end-user feedback via API - Automatic dataset curation from the Feedback Stream for fine-tuning pipelines - Support for DPO and RLHF-based model improvement using logged production data - Compatibility with agentic frameworks including LangChain and Magentic - Real-time accuracy signal and error detection for production LLM applications - Evaluation metrics including BLEU, ROUGE, strict and fuzzy matching
Additional Information
LLM-powered Agents Platform. Build, measure, deploy and improve your AI
Not public (pricing page exists at log10.io/pricing but specific tiers were not surfaced in search results)