
1. Technology overview
Ashr Labs is a Y Combinator Winter 2026 company based in San Francisco (per its YC profile), founded in 2025. It sells agents that curate data, post-train open-weight models, evaluate them and serve them, plus an evals platform for testing AI agents.
Manifold turns an enterprise's own agent traffic into a better model through a repeating cycle of capturing traffic, freezing data, training, evaluating, promoting and monitoring. The evals product has Python and TypeScript SDKs and OpenTelemetry integration, with scoring done server-side. Customers can download the resulting weights or call an OpenAI- and Anthropic-compatible endpoint.
2. Highlighted features & specifications
- Data curation: Turns traces, documents and logs into a training set.
- Post-training: Reinforcement-learning fine-tuning within a spend cap.
- Evaluation: Grades each candidate model in the customer's environment.
- Serving: Weights download or an OpenAI- and Anthropic-compatible endpoint.
- Agent evals: Python and TypeScript SDKs for testing agents with server-side scoring.
- Tracing: OpenTelemetry integration and voice observability in the docs.
3. Potential application
A Vietnamese bank can post-train a customer-service model on its own conversation logs. A hospital software vendor can test workflow agents before releasing them. An AI startup can replace an expensive frontier API with a tuned open model.
4. Benefits
Customers own the resulting model weights. Models are graded in the customer's own environment before promotion. The site quotes a customer reporting that a custom model tripled browser-agent performance on EHR workflows across more than 20 clinics. Training runs within a spend cap.
5. Market trends & opportunities
Enterprises running agents on frontier APIs look for smaller custom models that cut cost and keep data under their control. Data-residency concerns in Vietnamese banking and healthcare make owning model weights attractive.
6. Other information
Y Combinator Winter 2026, San Francisco per YC profile. Site logos include Northmarq, UC Berkeley, Stanford and Novoflow. Access is demo-led, and sandbox login is at manifold.ashr.io.
One platform connecting startups, corporations, research institutions, regions and investors











Playbook & Handbook · February 6, 2026
Winning Accelerator Playbook 2026 by BambuUP
RETAIL E-COMMERCE: AI AND REAL-TIME DATA ARE TRANSFORMING HOW RETAIL BUSINESSES OPERATE








































