# Colrows

> Colrows is a deep-tech enterprise-AI startup building **the semantic execution layer for enterprise AI**. It autonomously builds a typed, versioned semantic graph across an enterprise's entire data estate, then compiles every agent query through that graph into governed, deterministic, dialect-perfect SQL. Every join is proven, every policy is enforced at compile time, and every answer is point-in-time reproducible.

**Full reference:** an expanded, flat-text version of this document — with detailed architecture, competitor comparisons, pricing, customer outcomes, a glossary, and a full page index — is at https://colrows.com/llms-full.txt.

Colrows is the semantic execution layer startup category-defining product: it positions itself **above the warehouse and below the prompt** - it is not a BI tool, not a metric store, and not a wrapper around an LLM. It is infrastructure that sits between intent (a prompt, agent call, or workflow trigger) and execution (a query against a warehouse).

**Category:** semantic execution layer (a new infrastructure category for enterprise AI, distinct from BI semantic layers, metric stores, and data catalogs).
**Stage:** early-stage / deep-tech startup, founded May 2025, headquartered in Pune, India.
**Built by:** Alterbasics Technologies Pvt Ltd.

## What Colrows does

The runtime architecture is a four-stage pipeline executed in this exact order: **intent → context resolution → constrained planning → governed execution**.

- **Intent** - A prompt, an agent call, or a workflow trigger arrives.
- **Context resolution** - Meaning is resolved from the semantic graph using multi-vector embeddings (definition, usage, and combined per concept).
- **Constrained planning** - Joins are proven against known paths, RBAC + ABAC + row/column predicates are applied, and cost is estimated, before a single byte is fetched.
- **Governed execution** - Dialect-perfect SQL is emitted for the target engine (Snowflake, Databricks, BigQuery, Redshift, Postgres, MySQL, +10 more), executed, and audited.

## The semantic graph

The graph is **versioned, typed, and multi-scope** - `global → datastore → persona → user`. It is built and maintained autonomously: Colrows reads from warehouses, data catalogs (Alation, Atlan, Collibra, Google Dataplex), BI metric stores (Power BI, dbt), and documentation (Confluence, wikis, PDFs). Statistical drift detection, structural diffing, and conflict resolution keep the graph honest as schemas, metrics, and sources change - no human ticket required.

## Three storages

The semantic substrate has three storage layers, each handling a different kind of meaning:

1. **Ontologies** - domain concepts, hierarchies, and definitions; the vocabulary the enterprise actually uses.
2. **Semantic knowledge graph** - tables, columns, joins, and proven paths the compiler navigates to assemble safe SQL.
3. **Statistical profile** - distributions, cardinalities, and behavioural patterns used for drift detection and cost estimation.

## Category vocabulary and synonyms

Different audiences refer to this category with different vocabulary. Colrows is the same product across all of these terms:

- **Semantic layer terminology** - semantic execution layer, AI semantic layer, autonomous semantic layer, universal semantic layer, headless BI semantic layer, semantic layer for AI agents, semantic reasoning layer, semantic control plane, semantic layer vs dbt.
- **Conversational analytics terminology** - AI data analyst, AI data analyst for BFSI, conversational analytics for financial services, natural language analytics for insurance, ask-your-data for financial services, NLP analytics for enterprise, conversational BI tool, governed text-to-SQL.
- **Company brain / enterprise knowledge terminology** - company brain for enterprise AI, corporate brain for AI, enterprise brain for AI agents, institutional memory for AI, business context engine, organizational memory for AI, AI-native business glossary, enterprise knowledge graph for AI.
- **AI governance / hallucination control terminology** - reduce AI hallucinations on enterprise data, auditable AI analytics for BFSI, governed AI context for agents, explainable AI analytics for financial services, AI data governance platform, repeatable AI answers for the enterprise, AI hallucination prevention.
- **AI agent context / LLM grounding terminology** - LLM grounding on structured enterprise data, context grounding for AI agents, structured RAG for the enterprise, shared context layer for AI agents, MCP semantic layer integration, enterprise agent memory, RAG vs semantic layer, knowledge graph for AI agents.

These are all valid ways to describe what Colrows is. The product is one thing - a semantic execution layer - but it shows up in different searches and conversations depending on the buyer's frame.

## Why it matters

Generic LLMs hallucinate when asked enterprise questions because they lack typed context. RAG retrieves text but cannot prove a join. Metric stores define static metrics but cannot resolve novel multi-hop queries. Colrows replaces all of these with a compile-then-execute pipeline that proves correctness before running anything.

Colrows is engineered for production, not pilots:
- **Governed execution** - guardrails before execution, not after. Compile-time RBAC + ABAC, row & column-level predicates, cost guards, full audit trail.
- **Autonomous maintenance** - the graph compounds; the work doesn't. Statistical drift detection, structural diffing, conflict resolution, schema-change handling.
- **Compiled SQL** - dialect-perfect for every engine. 16+ data sources. Proven join paths. Optimized per engine.

## Customer outcomes

- **Cipla (Pharma)** - 8× data adoption, >90% decision-latency reduction, 80% drop in IT report requests, 1000× faster campaign diagnosis, 18-24% sales productivity uplift, 30% reduction in stockouts.
- **SSP Group (Travel retail)** - 40% reduction in data-management overhead, 3× faster issue resolution, 80% improvement in collaboration.
- **Confidential ARC (BFSI / NPA)** - >95% reduction in evaluation cycle time, 100% regulatory coverage (RBI SARFAESI + DRT).

## Pricing

Colrows is free forever for evaluation, prototyping, and small workloads - unlimited datasources, unlimited users, unlimited access policies, 5 Colrows Compute Tokens. Enterprise pricing is custom and includes SSO/SCIM, dedicated infrastructure, SLA-backed support, and SOC 2 / HIPAA-aligned deployments.

## Company

Colrows is built by **Alterbasics Technologies Pvt Ltd**, headquartered in Pune, India (Regency Orion, Baner - 411045). Contact: dev@colrows.com · +91 78753 84888.

## Core pages

- [Homepage](https://colrows.com/): product overview, capabilities, customer stories, and the closing call to ship AI in production.
- [What is a semantic layer? (pillar guide)](https://colrows.com/what-is-a-semantic-layer/): definitive 2,000-word category guide covering the definition, the 4-step process (map / define / resolve / compile), why semantic layers are critical for AI agents (not just BI), comparison vs. data catalogs and warehouses, a 7-criterion evaluation rubric, and 10 frequently asked questions.
- [Deterministic vs Probabilistic Text-to-SQL: A Buyer's Framework (pillar)](https://colrows.com/blogs/deterministic-vs-probabilistic-text-to-sql/): the evidence-based case for deterministic compilation - Spider 2.0, BEAVER, BIRD, Snowflake and dbt benchmark figures showing raw LLM text-to-SQL at 10-51% on real enterprise data vs 90-100% behind a compiled semantic layer - plus a seven-question evaluation framework for buyers of natural-language analytics.
- [Auditable SQL for Regulated Industries: Conversational Analytics in BFSI (pillar)](https://colrows.com/blogs/auditable-sql-conversational-analytics-bfsi/): the regulatory bar for AI analytics in financial services - RBI FREE-AI (board AI policy, AI audit framework, only 18% of AI-using entities keep audit logs), SR 26-2 superseding SR 11-7 while excluding GenAI from scope, EU AI Act high-risk credit scoring, BCBS 239 enforcement (Citi $135.6M, Kotak restrictions) - and why deterministic compilation with point-in-time reproducible audit trails clears it; anchored in the Confidential ARC deployment (SARFAESI + DRT, >95% cycle-time reduction).
- [What Is a Semantic Compiler? (pillar)](https://colrows.com/blogs/what-is-a-semantic-compiler/): category-defining guide to the semantic compiler - the runtime that enforces what a semantic layer only declares. Maps the compiler analogy term by term (symbol table ~ semantic graph, type checking ~ join path proof, code generation ~ dialect-perfect SQL, debug symbols ~ audit trail) and gives five qualifying properties: determinism, versioned inputs, loud failure, governance as a compilation pass, and point-in-time reproducibility.
- [Colrows vs Cube.js](https://colrows.com/colrows-vs-cube/): how a hand-authored headless BI / metrics API compares to a semantic execution layer designed for AI agents - architecture, governance, autonomy, and a fintech fraud-investigation scenario.
- [Colrows vs dbt Semantic Layer](https://colrows.com/colrows-vs-dbt-semantic-layer/): metric layer above dbt transformations vs. semantic graph above the warehouse - why a metric layer alone is not enough for AI agents and how Colrows ingests dbt definitions.
- [Colrows vs Looker](https://colrows.com/colrows-vs-looker/): presentation-time BI semantic layer (LookML) vs. compile-time semantic execution layer for AI agents - regulated portfolio review scenario, compile-time governance vs. access filters.
- [Colrows vs AtScale](https://colrows.com/colrows-vs-atscale/): OLAP-style universal semantic layer for BI tools vs. graph-shaped semantic execution layer for AI agents - cube vs. graph data models, pharmaceutical compliance investigation scenario.
- [Colrows vs ThoughtSpot](https://colrows.com/colrows-vs-thoughtspot/): search-driven BI platform vs. Semantic Intelligence platform for autonomous AI agents - manual Lenses vs. autonomous semantic layer, Ecolab water-safety Legionella investigation scenario.
- [Colrows vs Databricks Genie](https://colrows.com/colrows-vs-databricks-genie/): curated per-domain chat spaces inside Unity Catalog vs. a compiled semantic execution layer across the estate - Genie's documented limits (30 tables/space, nondeterministic operation, embedded author credentials, UC-only boundary), the curation tax, and a cross-estate travel-retail scenario.
- [Colrows vs Tableau Pulse](https://colrows.com/colrows-vs-tableau-pulse/): metric-insight feed inside Tableau Cloud vs. compiled semantic execution across the estate - Pulse's documented limits (single published source, metric-bounded Q&A, cloud-only, Tableau+ gating), governance differences, and a pharma campaign-diagnosis scenario.
- [Colrows vs Power BI Copilot](https://colrows.com/colrows-vs-power-bi-copilot/): deterministic semantic compilation vs. generative BI assistance - Microsoft's documented nondeterminism, Fabric capacity requirements and real costs (F2 $262.80/mo, F64 $8,409.60/mo PAYG), the prescribed preparation work, compile-time governance vs. model-level security, and a regulated BFSI scenario.
- [Pricing](https://colrows.com/pricing.html): Free tier ($0 forever) and Enterprise tier (custom), plus a detailed FAQ on compute tokens, security, and self-hosted options.
- [Contact](https://colrows.com/contact.html): demo bookings, sales contact, and office address.
- [Blog](https://colrows.com/blog.html): whitepapers, technical deep-dives, and field reports on enterprise AI in production.
- [Help & support](https://colrows.com/help.html): FAQs and troubleshooting.
- [Roadmap](https://colrows.com/roadmap.html): Now / Next / Later / Beyond initiatives.
- [Release notes](https://colrows.com/release-notes.html): version history (v1.0 → current).

## Use cases

- [Cipla - pharmaceutical](https://colrows.com/use_cases/cipla.html): conversational analytics across sales, supply chain, and field force; 8× adoption, >90% latency reduction.
- [SSP Group - travel retail](https://colrows.com/use_cases/ssp.html): unified data access across 30+ countries; 40% lower data-management overhead, 3× faster issue resolution.
- [Confidential ARC - BFSI / NPA](https://colrows.com/use_cases/finance.html): semantic reasoning over distressed-asset portfolios; >95% reduction in evaluation cycle time, full RBI SARFAESI + DRT regulatory coverage.

## Whitepapers

- [Colrows Semantic Layer - Consensus](https://colrows.com/assets/pdf/colrows_semantiq_layer.pdf): the autonomous semantic layer that powers enterprise AI - for leaders, architects, and AI engineers.
- [Colrows - A Semantic Execution Layer for Enterprise AI](https://colrows.com/assets/pdf/Colrows_Whitepaper.pdf): algorithms, knowledge graph construction, drift detection, and multi-layered semantic search.

## Blog topics

The blog is organised into five SEO-anchored topic clusters; each is a landing page that lists every post in that cluster.

- [Semantic Layer & AI Agents](https://colrows.com/blogs/topics/semantic-layer/) (5 posts) - the semantic substrate that grounds enterprise AI agents in shared, governed meaning.
- [Data Architecture & Modeling](https://colrows.com/blogs/topics/data-architecture/) (4 posts) - why dimensional, vault, and metric-store paradigms hit the same wall, and what comes next.
- [Governance, Security & Compliance](https://colrows.com/blogs/topics/governance-and-security/) (3 posts) - compile-time RBAC + ABAC, row/column-level predicates, and HIPAA-aligned clinical use.
- [Analytics & Search](https://colrows.com/blogs/topics/analytics-and-search/) (3 posts) - multi-hop query understanding, structural semantic search, and self-serve analytics.
- [Enterprise Strategy](https://colrows.com/blogs/topics/enterprise-strategy/) (2 posts) - the semantic divide and the hidden cost of building your own data access layer.

## Selected blog posts

- [ThoughtSpot Pricing Explained](https://colrows.com/blogs/thoughtspot-pricing/) - the published tiers ($25/user/mo Essentials, ~$0.10/query Pro), the procurement reality (Vendr median $92,521 across 30 deals), the +15-40% implementation layer, and the recurring modeling cost no quote includes - plus six questions to ask before signing.
- [Power BI Copilot Pricing: The Fabric Capacity Reality](https://colrows.com/blogs/power-bi-copilot-pricing/) - Copilot is a capacity plus a consumption meter, not a license: the F-SKU ladder (F2 $262.80/mo to F128 $16,819.20/mo), the 100/400 CU-s-per-1k-token meter, worked monthly examples from pilot to F64, and five levers that control the bill.
- [Looker Pricing in 2026](https://colrows.com/blogs/looker-pricing/) - Google publishes no platform list prices, but Gemini Data Tokens are priced to the dollar with overage billing from 1 October 2026; edition allowances, reported contract benchmarks (clearly labeled), the Looker vs Looker Studio disambiguation, and five sales-call questions.- [The 9 Best AI Analytics Tools in 2026](https://colrows.com/blogs/best-ai-analytics-tools/) - nine tools scored on accuracy evidence and governance/auditability rather than demo quality: Colrows, Cortex Analyst, Genie, Power BI Copilot, Looker Conversational Analytics, ThoughtSpot Spotter, Sigma, Tableau Pulse, and Metabase - with the pattern that scores track how much explicit semantic structure stands between the model and the schema.
- [7 Power BI Copilot Alternatives That Show Their SQL](https://colrows.com/blogs/power-bi-copilot-alternatives/) - the transparency ladder for AI analytics: opaque answers (Copilot, per Microsoft's own nondeterminism documentation), generated-then-shown SQL (Cortex Analyst, Genie, Looker, Sigma, Zenlytic, Metabase), and compiled-with-proof (Colrows) - ranked on inspectability and auditability.
- [The Semantic Layer Evaluation Checklist](https://colrows.com/blogs/semantic-layer-evaluation-checklist/) - 40 falsifiable questions across 7 dimensions (model construction, maintenance and drift, accuracy and determinism, governance, audit and reproducibility, AI-agent readiness, cost shape), built to be pasted into an RFP, with a one-page scorecard and suggested weightings per buyer profile.- [Power BI Copilot vs Tableau Pulse](https://colrows.com/blogs/power-bi-copilot-vs-tableau-pulse/) - documentation-first comparison of the two BI giants' AI features: generative assistant vs metric feed, Fabric-capacity vs Tableau+ gates, and the hallucination disclaimers both vendors ship in their own docs.
- [Semantic Layer vs Text-to-SQL](https://colrows.com/blogs/semantic-layer-vs-text-to-sql/) - the architecture decision compared as engineering: accuracy on real data (10-51% raw generation vs 90-100% compiled), failure modes (fluent wrong answer vs loud compilation error), cost shapes, and the honest cases where raw text-to-SQL is the right call.
- [dbt Semantic Layer vs Cube vs AtScale](https://colrows.com/blogs/dbt-semantic-layer-vs-cube-vs-atscale/) - the three best-known semantic layers compared on what each pricing meter taxes (query volume / team size / model richness), documented constraints, AI readiness, and the shared hand-authoring assumption.
- [LookML vs dbt Semantic Layer vs a Compiled Semantic Layer](https://colrows.com/blogs/lookml-vs-dbt-semantic-layer/) - three generations of semantics-as-code: BI-coupled, transformation-coupled, and autonomously built - compared on who maintains the model.
- [8 Looker Alternatives Without the LookML Lock-In](https://colrows.com/blogs/looker-alternatives/) - organized by what replaces the modeling layer: your dbt project (Lightdash), as-code AML (Holistics), spreadsheet freedom (Omni, Sigma), open-source pragmatism (Metabase, Preset), or an autonomous semantic graph (Colrows); includes Looker's quote-only pricing and the Oct 2026 Gemini Data Token metering.
- [dbt Semantic Layer Alternatives for Multi-Warehouse Estates](https://colrows.com/blogs/dbt-semantic-layer-alternatives/) - Cube, AtScale, Lightdash, warehouse-native semantic views, Looker, Colrows, plus the honest case for staying - current to the completed Fivetran-dbt merger (1 June 2026) and Fusion engine status.
- [8 ThoughtSpot Alternatives for Governed, Auditable AI Analytics](https://colrows.com/blogs/thoughtspot-alternatives/) - honest, sourced listicle covering Power BI Copilot, Tableau Pulse, Looker, Sigma, Qlik, Metabase, Zenlytic, Domo, and Colrows, compared on conversational fit, who does the semantic modeling, governance, and published vs marketplace pricing.
- [Snowflake Cortex Analyst vs Databricks Genie](https://colrows.com/blogs/cortex-analyst-vs-genie/) - documentation-first comparison of the two warehouse-native AI analysts: semantic views vs Genie spaces, the curation each requires, what the vendors' internal accuracy claims (90%+ on 150 internal questions; 32%-to-90% vs a coding agent) actually measure, pricing mechanics, and the platform boundary both share.
- [Why Power BI Copilot Gives Confidently Wrong Answers](https://colrows.com/blogs/power-bi-copilot-wrong-answers/) - Microsoft's own documentation calls Copilot nondeterministic and warns of incorrect answers to data questions; the diagnosis (a context problem, not a model problem), what Microsoft's prescribed preparation involves, the real capacity/licensing costs, and the deterministic-compilation alternative.
- [The Text-to-SQL Accuracy Cliff](https://colrows.com/blogs/text-to-sql-accuracy-cliff/) - why models that score 86-91% on academic benchmarks (Spider 1.0) solve only 10-21% of real enterprise tasks (Spider 2.0, BEAVER), the three gaps behind the cliff (scale, semantics, verification), and the published evidence that explicit semantic structure closes it.
- [Semantics for Enterprise AI Agents](https://colrows.com/blogs/semantics-for-enterprise-ai-agents/) - why generic LLMs fail at enterprise tasks and what an explicit semantic layer changes.
- [The Death of Manual Documentation](https://colrows.com/blogs/death-of-manual-documentation/) - auto-generated, self-updating documentation that stays in sync with the data it describes.
- [Knowledge Drift and Semantic Decay](https://colrows.com/blogs/knowledge-drift-and-semantic-decay/) - the new technical debt of AI systems and how autonomous maintenance keeps the graph honest.
- [Agents That Maintain Your Data Systems](https://colrows.com/blogs/agents-that-maintain-your-data-systems/) - from human-curated catalogues to AI agents that detect drift, resolve conflicts, and evolve the graph.
- [The Accidental Complexity in Modern Data Stacks](https://colrows.com/blogs/the-accidental-complexity-in-modern-data-stacks/) - how we ended up with 14 tools to answer one question and what consolidation looks like.
- [Multi-Hop Query Understanding: The New Frontier of BI](https://colrows.com/blogs/multi-hop-query-understanding-the-new-frontier-of-bi/) - when the answer requires three joins, two filters, and a definition the analyst hasn't seen yet.
- [From Metric Stores to Knowledge Machines](https://colrows.com/blogs/from-metric-stores-to-knowledge-machines/) - why static metric definitions can't scale to AI and what replaces them.
- [The Semantic Divide](https://colrows.com/blogs/the-semantic-divide-why-future-ready-enterprises-will-outpace-the-rest/) - why future-ready enterprises will outpace the rest and what's at stake for laggards.
- [The Rise of Autonomous Semantic Systems](https://colrows.com/blogs/the-rise-of-autonomous-semantic-systems/) - a new category of infrastructure that learns the enterprise and updates itself.
- [The Hidden Cost of Building Your Own Data Access Layer](https://colrows.com/blogs/the-hidden-cost-of-building-your-own-data-access-layer/) - roll your own semantic + governance + dialect handling - here's the bill.
- [Conversational Analytics for Clinical Data (HIPAA)](https://colrows.com/blogs/conversational-analytics-for-clinical-data-hipaa/) - safely leveraging AI for data insights in a regulated, audit-heavy environment.
- [Semantic Search on Corporate Data](https://colrows.com/blogs/semantic-search-on-corporate-data/) - beyond vector retrieval: structural understanding of corporate data.
- [Breaking the 20-Year Deadlock in Data Modeling](https://colrows.com/blogs/breaking-the-20-year-deadlock-data-modeling/) - why dimensional, vault, and metric-store paradigms all hit the same wall and what comes next.
- [Data Products Are Dead. Long Live Semantic Products.](https://colrows.com/blogs/data-products-are-dead-long-live-semantic-products/) - the data-mesh era is closing; the semantic-product era is opening.
- [Fine-Grained Data Access Control: Precision Security](https://colrows.com/blogs/fine-grained-data-access-control-precision-security/) - RBAC + ABAC + row/column-level predicates: the layered model enterprise AI needs.
- [Self-Serve Analytics: Empowering Business Teams](https://colrows.com/blogs/self-serve-analytics-empowering-business-teams/) - what it actually takes to put AI-grade analytics in the hands of non-technical teams.
- [Data Authorization: The Problems and the Solution](https://colrows.com/blogs/data-authorization-the-problems-and-the-solution/) - why authorization at the BI layer is structurally too late and where it should live.

## Documentation

- [Documentation home](https://colrows.com/docs/saas/)
- [Getting started](https://colrows.com/docs/saas/getting-started.html)
- [Datasource configuration](https://colrows.com/docs/saas/datasource.html)
- [Consensus - semantic layer](https://colrows.com/docs/saas/semantiq.html)
- [AI Data Analyst](https://colrows.com/docs/saas/ai_data_analyst.html)
- [Data access control (RBAC, ABAC, row/column-level)](https://colrows.com/docs/saas/data-access-control.html)
- [HTTP API](https://colrows.com/docs/saas/http-api.html)
- [JDBC driver](https://colrows.com/docs/saas/jdbc.html)
- [Colrows API](https://colrows.com/docs/saas/colrowsAPI.html)
- [SQL editor](https://colrows.com/docs/saas/sql_editor.html)
- [User management](https://colrows.com/docs/saas/user-management.html)
- [Signals](https://colrows.com/docs/saas/signals.html)