# SEAM - Semantic Engine for Agent Mediation > SEAM is the intelligence layer between AI agents and the data sources they query. Built by Measurelab. It ensures every agent answer is consistent, governed and auditable - without moving or replatforming any data. ## The situation AI agents now query live business systems directly - CRMs, billing platforms, databases, documents, Slack, email. The friction that justified centralised data warehouses as the access layer has collapsed. People in your organisation are already wiring agents into HubSpot, BigQuery, Jira. They didn't ask. They went where the friction was lowest. That speed comes with a cost: when different agents query different systems with different definitions, the same question produces different answers - with no audit trail and no shared notion of what the answer means. ## What SEAM is SEAM is the governed access layer for AI-agent traffic. It: - Resolves natural-language questions against governed metric, entity and policy definitions before any query executes - Maintains a single source of truth for what business terms mean - version-controlled, reviewed and owned like code - Provides full audit trails for every agent interaction - Works across structured and unstructured sources (databases, APIs, documents, Slack, email) - Requires no data movement or replatforming - governs data wherever it already lives - Plugs into existing agent stacks via the Model Context Protocol (MCP) and other open interfaces ## SEAM Canvas SEAM Canvas is the authoring surface for an organisation's intelligence layer. More than a glossary: it captures the full reasoning context - metrics, entities, policies, source hierarchies, fiscal-year boundaries, attribution rules, segmentation logic and how the people who run the business actually describe what they mean. The Canvas is the artefact teams build collaboratively. SEAM is the engine that runs from it. ## Architecture SEAM consists of: - **Definition files (YAML)** - version-controlled schemas defining tools, metrics, entities and access policies - **Compiler / validator** - validates definitions against schemas before deployment - **Context resolver** - maps agent intent to the correct governed definition - **Agent middleware** - plugs into existing agent stacks via MCP and other interfaces - **Audit logger** - full execution trace of every prompt, call and result Built on open standards. Extended where they stop. ## Key concepts - **Metric definitions** - what terms like "revenue", "active customer" or "churn rate" mean in a specific business context, with team-specific variants - **Source hierarchy** - rules for which system takes precedence when systems disagree - **Entity resolution** - how records in one system map to records in another - **Business logic** - fiscal-year boundaries, attribution models, segmentation rules - **Temporal context** - version history of definitions, so the right definition applies to the right time period ## What SEAM is not - Not a centralised data platform - it doesn't move or store your data - Not a glossary - definitions are executable and enforced, not descriptive - Not a replacement for a data warehouse - it complements existing infrastructure - Not a model - it sits between models and data, not in place of either ## SEAM and agentic analytics Agentic analytics - autonomous AI agents that query, reason over and act on your data - is only as reliable as the semantic layer beneath it. SEAM is that layer: governed metric and entity definitions that give agents verified meaning instead of guesses. It is how you make agentic analytics trustworthy. Built on open standards (MCP). SEAM is the readiness and governance layer - not an agent itself. ## Partner programme SEAM is in active development. Organisations can join the partner programme at three levels: - **Early access** - hands-on deployment with direct engineering support - **Co-development** - joint governance schema development with custom connectors - **Design partner** - deep integration with preferential pricing at general availability ## Benchmark A controlled head-to-head bench (Measurelab whitepaper, April 2026) measured what an intelligence model does to an AI agent meeting your data. SEAM (governed) versus Direct (raw MCP connectors only), both running claude-sonnet-4-6 at temperature 0, across 14 tasks spanning ten Measurelab connection types (BigQuery, GA4, GTM, Slack, Drive, GitHub, GCP, OpsHub, Harvest, Atlassian). Headline findings on 39 paired trials: - Correctness: 95% (SEAM) vs 44% (Direct), +51 pp [+36, +67] - Caveat capture: 46% vs 8%, +39 pp [+23, +54] - Cost per pass: $14.38 vs $33.42 (Direct 2.3x more expensive) - Wall-clock latency: 476s vs 734s (Direct 1.5x slower) - Tool calls: 79 vs 121 (Direct 1.5x more) On defined-metric questions specifically (the hardest subset): 86% correct vs 21%, 93% caveat capture vs 21%. On entity-routing tasks: 100% correct vs 50%. Conclusion: governance trades nothing on speed and cost, and improves correctness. Full report PDF: https://measurelab.ai/assets/SEAM-vs-Direct-bench.pdf ## Contact - Email: hello@measurelab.co.uk - Website: https://measurelab.ai - Parent company: https://www.measurelab.co.uk ## Links - [Home - your people found a faster way](https://measurelab.ai/) - [Product - why, what, how](https://measurelab.ai/seam/) - [Use cases - seven scenarios where SEAM applies](https://measurelab.ai/seam/use-cases/) - [Pricing - Framework (free, open source on npm), Standard at £900/mo or £750/mo billed annually for up to 7 connectors and unlimited team members, Scale on custom pricing for more than 7 connectors and bespoke connector builds](https://measurelab.ai/seam/pricing/) - [Get involved - founding customer, design partner, implementation partner](https://measurelab.ai/seam/get-involved/) - [Manifesto - siloed data, central intelligence](https://measurelab.ai/seam/manifesto/) - [About Measurelab - twelve years of analytics consulting. SEAM is what we learned, packaged.](https://measurelab.ai/about/) - [Security - certifications, controls and AI governance](https://measurelab.ai/security/) - [Docs - concepts, glossary and CLI install](https://measurelab.ai/seam/docs/) - [Docs concepts - intelligence layer, intelligence model, manifest, resolver, framework vs runtime split](https://measurelab.ai/seam/docs/concepts/) - [Docs glossary - canonical SEAM terminology, anchor-linkable per term](https://measurelab.ai/seam/docs/glossary/) - [Docs CLI - npm install -g @measurelab/seam-cli; init, validate, compile, resolve](https://measurelab.ai/seam/docs/cli/) - [Benchmark report (PDF) - SEAM vs Direct head-to-head, April 2026](https://measurelab.ai/assets/SEAM-vs-Direct-bench.pdf)