From a5cbe498a3a849afabe8455a341ab9097b96550c Mon Sep 17 00:00:00 2001 From: LiJung Chi Date: Tue, 17 Mar 2026 13:43:28 +0800 Subject: [PATCH 1/5] Update README with a16z context-layer framing --- README.md | 21 ++++++++++++++++++--- 1 file changed, 18 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index a8feb1a6f..ef2cf25f4 100644 --- a/README.md +++ b/README.md @@ -39,6 +39,8 @@ AI agents can already call tools, browse docs, and write code. What they still s Enterprise data is not just rows in a warehouse. It is definitions, metrics, relationships, permissions, lineage, and intent. An agent that can connect to PostgreSQL or Snowflake still does not know what "net revenue", "active customer", or "pipeline coverage" actually mean in your company. +This is not just our thesis. In [Your Data Agents Need Context](https://a16z.com/your-data-agents-need-context/), a16z argues that data agents break down when they only have connectivity and SQL generation, but lack business definitions, source-of-truth context, and the operational knowledge that explains how a company actually runs. +

without_wren_engine

@@ -59,6 +61,8 @@ This is the open source context engine for teams building the next generation of We believe the future of AI is not tool calling alone. It is context-rich systems where agents can reason, retrieve, plan, and act on top of a shared understanding of business reality. +The a16z post captures this shift well: the market is moving beyond text-to-SQL and toward a living context layer that combines semantic meaning, system structure, governance, and human refinement. + Wren Engine is our open source contribution to that future. It is the semantic and execution foundation beneath Wren AI, and it is designed to be useful well beyond a single product: @@ -78,9 +82,12 @@ Wren Engine turns business data into agent-usable context. At a high level: 1. You describe your business domain with Wren's semantic model and MDL. -2. Wren Engine analyzes intent, models, relationships, and access rules. -3. It plans and generates correct queries across your underlying data sources. +2. Wren Engine captures the context agents need: models, metrics, relationships, and access rules. +3. It analyzes intent and plans correct queries across your underlying data sources. 4. MCP clients and AI agents interact with that context through a clean interface. +5. Teams keep refining the model as business logic and systems evolve. + +This is the practical open source path from text-to-SQL toward context-aware data agents. That means your agent is no longer asking, "Which raw table should I query?" @@ -107,6 +114,8 @@ Use it to power experiences like: - code assistants that need real business context, not just schema dumps - internal AI tools that should be grounded in semantic models instead of ad hoc SQL +This is especially important in developer-facing agent environments, where the assistant may understand your codebase but still lacks the business context required to answer data questions correctly. + ## Why Open Source We think agent infrastructure should be composable. @@ -121,6 +130,8 @@ Wren Engine is open source so the community can: - build opinionated agent products on a transparent foundation - help define what a real context layer for AI should look like +We want that context layer to be inspectable, composable, and community-owned, not trapped inside a single proprietary interface. + ## Architecture At A Glance ```text @@ -139,6 +150,8 @@ Core ideas: - `ibis-server` provides the execution and connector-facing API layer - `mcp-server` makes Wren easy to use from MCP-compatible agents +That last point matters: context only helps agents when it is available at runtime. Wren is built to expose that layer over MCP and APIs. + ## Repository Map This repository contains the core engine modules: @@ -171,10 +184,10 @@ Current open source support includes connectors such as: - MySQL - Oracle - PostgreSQL +- Redshift - SQL Server - Snowflake - Trino -- Redshift See the connector API docs in the project documentation for the latest connection schemas and capabilities. @@ -199,6 +212,7 @@ The MCP server includes: - [What is semantics?](https://docs.getwren.ai/oss/engine/concept/what_is_semantics) - [What is Modeling Definition Language (MDL)?](https://docs.getwren.ai/oss/engine/concept/what_is_mdl) - [Benefits of Wren Engine with LLMs](https://docs.getwren.ai/oss/engine/concept/benefits_llm) +- [Your Data Agents Need Context](https://a16z.com/your-data-agents-need-context/) - [Powering Semantic SQL for AI Agents with Apache DataFusion](https://getwren.ai/post/powering-semantic-sql-for-ai-agents-with-apache-datafusion) ### Developer entry points @@ -239,5 +253,6 @@ If you are building with agents today, this is a great time to get involved. - Open a [GitHub issue](https://github.com/Canner/wren-engine/issues) - Explore [Wren AI](https://github.com/Canner/WrenAI) to see the broader product vision - Read our mission piece: [Fueling the Next Wave of AI Agents](https://getwren.ai/post/fueling-the-next-wave-of-ai-agents-building-the-foundation-for-future-mcp-clients-and-enterprise-data-access) +- Read the market thesis from a16z: [Your Data Agents Need Context](https://a16z.com/your-data-agents-need-context/) Wren Engine is for builders who believe AI needs better context, not just better prompts. From e479f61ca170807fc5eaa67334b3937f98e54f0c Mon Sep 17 00:00:00 2001 From: LiJung Chi Date: Tue, 17 Mar 2026 14:39:31 +0800 Subject: [PATCH 2/5] update link --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index ef2cf25f4..0143f9b66 100644 --- a/README.md +++ b/README.md @@ -209,7 +209,7 @@ The MCP server includes: ### Learn the concepts - [Quick start with Wren Engine](https://docs.getwren.ai/oss/engine/get_started/quickstart) -- [What is semantics?](https://docs.getwren.ai/oss/engine/concept/what_is_semantics) +- [What is context?](https://docs.getwren.ai/oss/engine/concept/what_is_context) - [What is Modeling Definition Language (MDL)?](https://docs.getwren.ai/oss/engine/concept/what_is_mdl) - [Benefits of Wren Engine with LLMs](https://docs.getwren.ai/oss/engine/concept/benefits_llm) - [Your Data Agents Need Context](https://a16z.com/your-data-agents-need-context/) From d06e136b9ab4ade0301971b4a031a51918c603b5 Mon Sep 17 00:00:00 2001 From: LiJung Chi Date: Tue, 17 Mar 2026 15:50:58 +0800 Subject: [PATCH 3/5] docs: clarify Wren Engine positioning against adjacent tools --- README.md | 35 +++++++++++++++++++++++++++++++++++ 1 file changed, 35 insertions(+) diff --git a/README.md b/README.md index 0143f9b66..0d6d4ca51 100644 --- a/README.md +++ b/README.md @@ -93,6 +93,41 @@ That means your agent is no longer asking, "Which raw table should I query?" It is asking, "Which business concept, metric, or governed slice of context do I need to answer this task correctly?" +## Where Wren Engine Fits Compared To Other Approaches + +Teams sometimes ask how Wren Engine differs from connecting an agent to a catalog service such as DataHub through MCP, or from other tools that already expose metadata, BI models, or text-to-SQL access. + +The short answer is: many tools help agents discover data or generate queries, while Wren helps agents reason over business meaning and execute governed data access through a semantic layer. + +| Approach people often compare | Examples | What it is great at | What is still typically left unresolved for the agent | How Wren Engine differs | +| --- | --- | --- | --- | --- | +| Data catalog service via MCP | DataHub, similar catalog and metadata platforms | Metadata discovery, lineage, ownership, documentation, asset search | Choosing trusted metrics, resolving joins correctly, and turning metadata into governed runtime query behavior | Wren adds semantic models, relationships, metrics, and runtime query planning | +| Database connector or schema browser | Direct warehouse/database MCP servers, JDBC-style access, raw schema inspection tools | Fast access to tables, columns, and SQL execution endpoints | Business meaning is mostly implicit, so agents still have to infer intent from physical schema | Wren gives agents a business layer above raw tables and columns | +| BI or semantic tooling | Semantic layers, metrics layers, BI modeling tools | Defining curated metrics and business-friendly entities for analytics consumers | Many are designed first for dashboards or analyst workflows, not as an open runtime context layer for MCP-native agents | Wren is designed to expose semantic context directly to agent workflows through MCP and APIs | +| Text-to-SQL or SQL agent tooling | NL-to-SQL assistants, generic SQL copilots, LLM agents with database tools | Turning questions into SQL quickly when schema is simple or well-known | Accuracy drops when business definitions, joins, governance, or ambiguity matter | Wren reduces guessing by grounding generation in explicit semantic definitions | +| Knowledge base or documentation retrieval | Wiki search, docs search, RAG over data docs | Retrieving written explanations, runbooks, definitions, and usage notes | Retrieved docs may be stale, inconsistent, or not executable at query time | Wren operationalizes business context so it can be used consistently during planning and execution | + +Another way to frame it: + +| Dimension | Many adjacent tools | Wren Engine | +| --- | --- | --- | +| Primary role | Discovery, retrieval, documentation, or raw access | Semantic modeling, query planning, and governed execution context | +| What the agent mainly sees | Tables, columns, lineage, tags, owners, descriptions, docs, or SQL endpoints | Models, metrics, relationships, business definitions, and access rules | +| Main question it answers | "What data exists, and how can I inspect or query it?" | "What business concept or trusted metric should I use to answer this question correctly?" | +| SQL generation | Often left to the agent or a generic LLM layer | Built around translating intent through a semantic layer into correct queries | +| Join logic | Agent often has to infer joins from schema, lineage, or examples | Relationships are modeled explicitly so joins are not guessed ad hoc | +| Metric consistency | Definitions may exist in docs or dashboards, but enforcement is indirect | Metrics are defined in the semantic model and reused consistently | +| Governance at query time | Often visible as metadata or policy hints | Designed to carry governed business context into runtime query planning | +| Runtime role in an agent workflow | Helps the agent find, read, or access candidate data assets | Serves as the context and execution layer the agent uses to answer questions | + +In practice, many teams may want multiple layers working together: + +- a catalog service to inventory and govern the data estate +- documentation and lineage tools to help humans understand the environment +- Wren Engine to turn that estate into a semantic, agent-ready context layer + +If you only give an agent metadata, docs, or raw database access, it may still need to guess which joins, filters, and metric definitions are actually correct. Wren exists to reduce that gap between discovery and trustworthy execution. + ## Built For Agent Builders Wren Engine is especially useful for the open source community building agent-native workflows in tools like: From 1f275fc32e073ba0b15c71014b25521926b813b8 Mon Sep 17 00:00:00 2001 From: LiJung Chi Date: Tue, 17 Mar 2026 16:55:44 +0800 Subject: [PATCH 4/5] Update README with simpler Wren comparison and quickstart links --- README.md | 54 +++++++++++++++++++++++++----------------------------- 1 file changed, 25 insertions(+), 29 deletions(-) diff --git a/README.md b/README.md index 0d6d4ca51..b7be5a3df 100644 --- a/README.md +++ b/README.md @@ -93,40 +93,35 @@ That means your agent is no longer asking, "Which raw table should I query?" It is asking, "Which business concept, metric, or governed slice of context do I need to answer this task correctly?" -## Where Wren Engine Fits Compared To Other Approaches +## Wren Engine vs. Other Data Tools -Teams sometimes ask how Wren Engine differs from connecting an agent to a catalog service such as DataHub through MCP, or from other tools that already expose metadata, BI models, or text-to-SQL access. +People often compare Wren Engine to catalog services like DataHub, raw database MCP servers, BI semantic tools, or text-to-SQL agents. -The short answer is: many tools help agents discover data or generate queries, while Wren helps agents reason over business meaning and execute governed data access through a semantic layer. +The simple difference is: -| Approach people often compare | Examples | What it is great at | What is still typically left unresolved for the agent | How Wren Engine differs | -| --- | --- | --- | --- | --- | -| Data catalog service via MCP | DataHub, similar catalog and metadata platforms | Metadata discovery, lineage, ownership, documentation, asset search | Choosing trusted metrics, resolving joins correctly, and turning metadata into governed runtime query behavior | Wren adds semantic models, relationships, metrics, and runtime query planning | -| Database connector or schema browser | Direct warehouse/database MCP servers, JDBC-style access, raw schema inspection tools | Fast access to tables, columns, and SQL execution endpoints | Business meaning is mostly implicit, so agents still have to infer intent from physical schema | Wren gives agents a business layer above raw tables and columns | -| BI or semantic tooling | Semantic layers, metrics layers, BI modeling tools | Defining curated metrics and business-friendly entities for analytics consumers | Many are designed first for dashboards or analyst workflows, not as an open runtime context layer for MCP-native agents | Wren is designed to expose semantic context directly to agent workflows through MCP and APIs | -| Text-to-SQL or SQL agent tooling | NL-to-SQL assistants, generic SQL copilots, LLM agents with database tools | Turning questions into SQL quickly when schema is simple or well-known | Accuracy drops when business definitions, joins, governance, or ambiguity matter | Wren reduces guessing by grounding generation in explicit semantic definitions | -| Knowledge base or documentation retrieval | Wiki search, docs search, RAG over data docs | Retrieving written explanations, runbooks, definitions, and usage notes | Retrieved docs may be stale, inconsistent, or not executable at query time | Wren operationalizes business context so it can be used consistently during planning and execution | +- those tools usually help an agent find data or generate SQL +- Wren helps an agent understand business meaning and produce the right query through a semantic layer -Another way to frame it: - -| Dimension | Many adjacent tools | Wren Engine | +| Tool type | What it gives the agent | What Wren Engine adds | | --- | --- | --- | -| Primary role | Discovery, retrieval, documentation, or raw access | Semantic modeling, query planning, and governed execution context | -| What the agent mainly sees | Tables, columns, lineage, tags, owners, descriptions, docs, or SQL endpoints | Models, metrics, relationships, business definitions, and access rules | -| Main question it answers | "What data exists, and how can I inspect or query it?" | "What business concept or trusted metric should I use to answer this question correctly?" | -| SQL generation | Often left to the agent or a generic LLM layer | Built around translating intent through a semantic layer into correct queries | -| Join logic | Agent often has to infer joins from schema, lineage, or examples | Relationships are modeled explicitly so joins are not guessed ad hoc | -| Metric consistency | Definitions may exist in docs or dashboards, but enforcement is indirect | Metrics are defined in the semantic model and reused consistently | -| Governance at query time | Often visible as metadata or policy hints | Designed to carry governed business context into runtime query planning | -| Runtime role in an agent workflow | Helps the agent find, read, or access candidate data assets | Serves as the context and execution layer the agent uses to answer questions | +| Data catalog services | Tables, columns, lineage, owners, descriptions | Business models, metrics, relationships, and governed query planning | +| Raw database or schema access | Direct access to schemas and SQL execution | A business layer above raw tables so the agent does not have to guess intent | +| BI or semantic tools | Curated metrics and entities for analytics workflows | An open context layer designed for MCP and agent workflows | +| Text-to-SQL agents | Fast SQL generation from natural language | Better accuracy by grounding generation in explicit business definitions | + +Many teams will want both: + +- a catalog to inventory and document the data estate +- Wren to turn that data into agent-ready context -In practice, many teams may want multiple layers working together: +Why that matters: -- a catalog service to inventory and govern the data estate -- documentation and lineage tools to help humans understand the environment -- Wren Engine to turn that estate into a semantic, agent-ready context layer +- more accurate answers because joins and metrics are defined instead of guessed +- more consistent answers because every agent uses the same business definitions +- safer data access because governance can be carried into query planning +- less prompt engineering because the context lives in the engine, not in the prompt -If you only give an agent metadata, docs, or raw database access, it may still need to guess which joins, filters, and metric definitions are actually correct. Wren exists to reduce that gap between discovery and trustworthy execution. +Without Wren, an agent may know where the data is but still not know how to answer the question correctly. ## Built For Agent Builders @@ -230,9 +225,11 @@ See the connector API docs in the project documentation for the latest connectio ### Use Wren through MCP -If you want to use Wren Engine from an AI agent or MCP-capable IDE, start here: +If you want to use Wren Engine from an Claude Code or MCP-capable IDE, start here: -- [MCP Server README](./mcp-server/README.md) +- [Quick start: Chat with jaffle_shop using Wren Engine + Claude Code](https://docs.getwren.ai/oss/engine/get_started/quickstart) +- [Quick start with Claude Desktop](https://docs.getwren.ai/oss/engine/get_started/quickstart_claude) +- [Understanding Wren AI project structure](https://docs.getwren.ai/oss/engine/get_started/structure) The MCP server includes: @@ -243,7 +240,6 @@ The MCP server includes: ### Learn the concepts -- [Quick start with Wren Engine](https://docs.getwren.ai/oss/engine/get_started/quickstart) - [What is context?](https://docs.getwren.ai/oss/engine/concept/what_is_context) - [What is Modeling Definition Language (MDL)?](https://docs.getwren.ai/oss/engine/concept/what_is_mdl) - [Benefits of Wren Engine with LLMs](https://docs.getwren.ai/oss/engine/concept/benefits_llm) From e16ef2211fb8f6ffe60e6e8133d1704f12d8a511 Mon Sep 17 00:00:00 2001 From: LiJung Chi Date: Tue, 17 Mar 2026 21:27:35 +0800 Subject: [PATCH 5/5] update text --- README.md | 16 +++++++--------- 1 file changed, 7 insertions(+), 9 deletions(-) diff --git a/README.md b/README.md index b7be5a3df..6c838a3ea 100644 --- a/README.md +++ b/README.md @@ -47,7 +47,7 @@ This is not just our thesis. In [Your Data Agents Need Context](https://a16z.com Wren Engine exists to solve that gap. -It gives AI agents a semantic layer they can reason over, so they can: +It gives AI agents a context layer they can reason over, so they can: - understand models instead of raw tables - use trusted metrics instead of inventing SQL @@ -59,9 +59,7 @@ This is the open source context engine for teams building the next generation of ## The Vision -We believe the future of AI is not tool calling alone. It is context-rich systems where agents can reason, retrieve, plan, and act on top of a shared understanding of business reality. - -The a16z post captures this shift well: the market is moving beyond text-to-SQL and toward a living context layer that combines semantic meaning, system structure, governance, and human refinement. +We believe the future of AI is not tool calling alone. It is context-rich systems where agents can reason, retrieve, plan, and act on top of a shared understanding of business reality. Moving beyond text-to-SQL and toward a living context layer that combines semantic meaning, system structure, governance, and human refinement. Wren Engine is our open source contribution to that future. @@ -100,9 +98,9 @@ People often compare Wren Engine to catalog services like DataHub, raw database The simple difference is: - those tools usually help an agent find data or generate SQL -- Wren helps an agent understand business meaning and produce the right query through a semantic layer +- Wren helps an agent understand business meaning and produce the right query through a context layer -| Tool type | What it gives the agent | What Wren Engine adds | +| Tool type | What it gives the agent | ***What Wren Engine adds*** | | --- | --- | --- | | Data catalog services | Tables, columns, lineage, owners, descriptions | Business models, metrics, relationships, and governed query planning | | Raw database or schema access | Direct access to schemas and SQL execution | A business layer above raw tables so the agent does not have to guess intent | @@ -112,7 +110,7 @@ The simple difference is: Many teams will want both: - a catalog to inventory and document the data estate -- Wren to turn that data into agent-ready context +- Wren Engine to turn that data into agent-ready context Why that matters: @@ -258,7 +256,7 @@ The MCP server includes: Common workflows: ```bash -# Rust semantic engine +# Rust context engine cd wren-core cargo check --all-targets @@ -274,7 +272,7 @@ cd mcp-server ## Project Status -Wren Engine is actively evolving in the open. The current focus is to make the semantic layer, execution path, and MCP integration stronger for real-world agent workflows. +Wren Engine is actively evolving in the open. The current focus is to make the context engine, execution path, and MCP integration stronger for real-world agent workflows. If you are building with agents today, this is a great time to get involved.