Myria

Myria is a standalone memory service for long-running agents.

It stores an append-only event log in PostgreSQL, builds versioned semantic snapshots in the background, and exposes the whole system through a stdio MCP server. The design target is a shared memory runtime that can serve conversations, planning, task execution, and group-chat settings without forcing clients to replay huge transcripts.

This project was inspired by PageIndex.

What Myria Does

• Appends canonical events with stable ingest-time IDs.
• Serves raw event trees through myria.query_event_nodes.
• Serves summary/reference trees through myria.query_references.
• Builds participant-scoped TopicIndex snapshots in the background.
• Keeps fresh post-snapshot events queryable while rebuilds are still pending.
• Exposes admin and observability tools for snapshots, topics, and build state.

What Myria Does Not Do

• It is not a general agent runtime.
• It is not a chat frontend.
• It does not expose an MCP network transport in v1.
• It does not trust emitter-supplied semantic IDs as canonical memory structure.

V1 Shape

• Language: Go
• Database: PostgreSQL
• Public interface: stdio MCP
• Internal LLM platform: OpenRouter
• Target model family for builder and tree-walker workflows: NVIDIA Nemotron 3 Super
• Internal logging: append-only file logging through service.log_file

Quick Start

1. Create a JSON config file for Myria.
2. Create an OpenRouter request template JSON.
3. Run migrations.
4. Start the stdio MCP server.

go run ./cmd/myria -config ./myria.json -cmd migrate
go run ./cmd/myria -config ./myria.json -cmd serve

Run a one-off build:

go run ./cmd/myria -config ./myria.json -cmd build

Run tests:

go test -timeout 5m ./...

Documentation

Start here:

• Docs Index
• Quick Start
• Configuration
• Architecture
• Data Model
• Indexing and Querying
• MCP API
• Storage
• Operations
• Full Design Specification

Suggested reading order:

1. Architecture
2. Quick Start
3. Configuration
4. Data Model
5. Indexing and Querying
6. MCP API
7. Operations
8. Full Design Specification

Project Layout

myria/
  cmd/myria/          entrypoint
  config/             sample LLM request template
  docs/               human-facing docs and full design spec
  internal/config/    config loading and validation
  internal/core/      shared types and ID logic
  internal/llm/       OpenRouter client and tool loop
  internal/mcp/       stdio MCP server
  internal/service/   orchestration, query paths, builder logic
  internal/store/     PostgreSQL schema and data access
  internal/testutil/  local test helpers

Current Status

Myria currently supports:

• stdio MCP serving
• PostgreSQL migrations and storage
• append/query/build flows
• inactivity and threshold-triggered builds
• exact event/topic lookup
• snapshot and builder observability
• live OpenRouter-backed LLM integration tests

The root README is the canonical entry point for the project. The full design lives in docs/design.md, and the rest of the docs are shorter guides organized for human readers.