From the source
Sarvam built a knowledge fabric that reads an organisation's documents in whatever form they arrive and turns them into a structured layer agents can navigate, query and cite.
- The company says the work happens once and is reused across questions, which makes agents more accurate and cheaper to run at scale.
- Sarvam says it refined the fabric with clients over the last few months and tested it on public benchmarks.
- On
Harvey's LAB benchmark, with GLM 5.3 as the agent model, answers met 84.2% of expert rubric criteria, up from 55.1% with agentic RAG, Sarvam says.
- The company says that result outperformed Harvey's own harness, reported at 79.1% for GPT-5.6 Sol and 71.7% for Opus 4.8.
- On Sierra's tau-knowledge banking benchmark, the fabric matched baseline accuracy with 64% fewer tokens, with GLM 5.2 running both setups.
- The fabric keeps its learning in two layers: a knowledge tree that organises a corpus so an agent can navigate it, and a domain ontology that models entities and their relationships.
- Sarvam serves the tree over MCP, so any agent can list folders, read indexes, search the corpus, open page ranges and inspect tables, with every citation pointing back to the source page.
