The Colrows platform
Colrows is the agentic platform for enterprise data. The consensus layer is built from your data estate, every question is grounded in it, and your agents run on your own compute. Three layers, one loop.
The consensus layer is built from your data estate. Every question is grounded in it. Agents run on your compute.
One loop. Feedback, curator decisions and the gaps agents report flow back into the consensus engine.
Colrows reads what your organization already knows, wherever it lives, and reconciles it into a single semantic layer that people and agents share.
The single definition of how your business is measured, and where every number comes from.
One source of truth for every agent. Analysts, curators and your own agents all read from this layer, and what they learn flows back into it.
Whether the answer lives in a warehouse, a document or a past decision, every question goes through the same semantic layer and the same access rules.
Agent REST and Agent MCP share the same tools, scopes and answers. Every call runs with the caller’s own access.
Questions resolve to governed SQL over approved metrics and verified queries, and run directly against your warehouses and databases.
Questions over documents, wikis and PDFs, answered in the language of the semantic layer and within each user’s document access.
Agents remember preferences, proven query patterns and the decisions people made, across sessions and runs.
Metrics, entities, terms and verified queries from the consensus layer.
Agents are registered, reviewed and seated in Colrows. Their runs execute on workers inside your own environment, with only the tools they were approved for.
Answers business questions with governed SQL and cited documents. Every finding carries the evidence that backs it.
Keeps the semantic layer healthy: works through open inquiries and proposals, and the gaps other agents report.
Same spec, same review, same guardrails, in the framework you choose.
The loop closes. Curator decisions and the gaps agents report flow back into the consensus layer, so every run makes the next one better. Back to the foundation →
Colrows Quorum is the agent control plane. You register an agent with its owner, spec, tools and semantic scope; an admin reviews and seats a version before it can run; runs start on demand or on a schedule within per-agent limits; every answer is checked against the evidence the run produced; and every run is audited: who ran what, as whom, with which tools and outcome.
On your compute. Colrows is the control plane; workers inside your environment claim runs and execute the agent with your code, your framework and your model keys. Each run gets a short-lived token limited to the tools that version was approved for, acting either as the agent or on behalf of a user. Raw SQL is never part of a run's grant. Tools call back to Colrows over Agent REST or Agent MCP.
Two. The AI Analyst answers business questions with governed SQL and cited documents, and every finding carries the evidence that backs it. The Curator keeps the semantic layer healthy: it works through open inquiries and proposals, and the gaps other agents report. Both are seated like any other agent. Your own agents follow the same spec, review and guardrails, in the framework you choose.
Every run result is checked against the evidence the run actually produced: the governed metrics, verified queries and cited documents it used. Because every data access goes through Colrows, the record of what each agent read is Colrows's own telemetry, not the agent's report of itself.
The part of Colrows that builds the semantic layer. It extracts signals from databases, data catalogs, analytics platforms, documents and user feedback; reconciles claims, so agreeing evidence builds confidence and conflicts are surfaced; proposes metrics, entities and terms with the evidence behind each; and keeps every curator decision so nobody is asked twice.
Docent is the document side of grounded access. It answers questions over documents, wikis and PDFs in the language of the semantic layer and within each user's document access. Every claim cites its section, answers stay current as documents change, and what the documents say feeds the consensus layer.
Yes. The semantic layer is the foundation: metrics, entities, business terms, relationships, verified queries and access policies, built and kept current by the consensus engine. Grounded access and Quorum are the layers that let people, applications and agents use it safely.
A 30-minute walk through the consensus layer, grounded access and a Quorum agent running on your compute.
