Meridian

OPEN SOURCE / APACHE-2.0

Code moves fast.
Give it a shared map.

An open-source context plane for the AI-driven development lifecycle (AI-DLC). Meridian connects architecture, contracts, ownership, and change impact so humans and coding agents work from a shared system map.

Explore the project See how it works

Built on MCP and AGENTS.md. Independent today; preparing for engagement with the AAIF community.

Structured knowledgeScoped accessCross-repo impactAgent-agnostic delivery

01 / THE PROBLEM

Faster code is not
the same as safer change.

Coding agents accelerate implementation. The hard part is knowing what a change means for the rest of the system.

01

No shared system map

Repositories are connected in production, but agents and developers often explore them in isolation.

02

Invisible blast radius

A local change can affect downstream consumers, API contracts, and services owned by another team.

03

Knowledge drift

Architecture diagrams, ownership records, and wiki pages can fall behind the code they describe.

04

Governance gaps

People and agents need context scoped to the right organization, portfolio, and application.

02 / THE SOLUTION

A living, governed
knowledge graph.

Declare architecture as structured facts. Compile a versioned system map. Deliver relevant context through the tools people and agents already use.

  1. 1

    Declare

    OKF YAML describes services, modules, contracts, datastores, teams, and conventions.

    Git + Open Knowledge Format
  2. 2

    Compile

    Validate facts and build relationships, impact sets, graph snapshots, and repository instructions.

    Versioned knowledge builds
  3. 3

    Connect

    Combine application knowledge at portfolio and organization scope to reveal cross-repo relationships.

    Organization / Portfolio / Application
  4. 4

    Use

    Agents query MCP tools; people explore the same knowledge through GraphQL and the web interface.

    MCP + GraphQL + AGENTS.md

CONTEXT AT THE POINT OF CHANGE

Ask better questions.
Start in the right place.

Use module context to locate relevant ownership and conventions. Use impact queries to identify affected consumers. Read the implementation and verify the change.

What module owns this file?

What depends on this contract?

Which conventions apply here?

What context was this agent given?

03 / THE VALUE

Less rediscovery.
More informed development.

DEVELOPMENT

Make change impact visible

Give reviewers and agents a shared view of touched modules, dependent services, contracts, and owners. PR advice informs review; it is not a blocking safety gate.

ONBOARDING & AUDIT

Make context traceable

Explore structured architecture instead of piecing together disconnected documents. Versioned builds and audit records show the source of served context.

AI-DRIVEN DEVELOPMENT

Measure the benefit

The evaluation harness compares tasks with and without Meridian context, tracking correctness, time, token usage when reported, and change scope.

THE AAIF ECOSYSTEM

Open protocols.
Shared system knowledge.

The Agentic AI Foundation (AAIF) hosts MCP, AGENTS.md, Goose, and Agentgateway. Meridian aims to complement these projects with versioned architecture knowledge, scoped context, impact analysis, and an audit trail.

AVAILABLE

MCP

Live tools serve module context, contracts, owners, conventions, and impact from Meridian's compiled knowledge.

Context delivery protocol
AVAILABLE

AGENTS.md

Generated repository instructions give file-reading coding agents a portable map of services, modules, and conventions.

Repository guidance format
PLANNED

Agentgateway

Planned as the MCP front door for policy-controlled connectivity and observability. Meridian remains the source of architecture context.

Connectivity and policy layer
VALIDATION PLANNED

Goose

An evaluation adapter exists. End-to-end Goose integration and real A/B runs with and without Meridian context are planned, not yet demonstrated.

Agent execution and evaluation

COMPLEMENT, NOT REPLACE

The missing map
between tools and code.

MCP defines how agents call tools. AGENTS.md provides repository guidance. Agentgateway connects and governs traffic. Goose executes agent tasks. Meridian supplies the system knowledge those workflows can use.

OKF: structured architecture facts in git

Meridian: compile, scope, trace, and query

MCP / AGENTS.md: deliver relevant context

Agents: inspect, implement, and verify

COMMUNITY ROADMAP

Build interoperability.
Publish the evidence.

Priorities for contributors and early adopters, not claims of completed integration.

CONNECT

Agentgateway integration

Document and validate routing to Meridian's MCP server, preserving tenant boundaries and testing authentication and observability end to end.

EVALUATE

Goose benchmarks

Validate the existing adapter with live runs. Publish reproducible same-model comparisons of task success, time, reported token usage, and change scope.

COLLABORATE

AAIF community engagement

Gather feedback on the knowledge format and integrations, improve contributor documentation, and prepare a project proposal through AAIF's published process.

Explore the proposal process

OPEN SOURCE / BUILT WITH GITHUB COPILOT

Explore it. Question it.
Help shape what comes next.

Meridian is open source under the Apache License 2.0. Built collaboratively with GitHub Copilot, it welcomes contributors across the agentic ecosystem: try the MCP tools, review OKF, help integrate Agentgateway, or validate Goose benchmarks.

View source Discuss an idea

Explore the demo estate

A connected commerce example: structured knowledge plus three application repositories.

Knowledgemeridian-okf-shop Frontendmeridian-storefront Ordersmeridian-orders Paymentsmeridian-payments