Skip to content

The Mercury Family

Advanced software foundations for human–AI collaboration — Accenture open source, Apache-2.0.

Mercury is a family of open-source foundations for building software with AI, not merely faster: composable engines whose applications are assembled from self-contained functions and YAML-configured event flows — a shape that human developers and AI agents read equally well — plus a shared memory layer that keeps every human and every AI agent on a project oriented on the same intent and state.

Members

Member Role Repository Documentation
agent-memory Vendor-neutral shared AI memory + cognitive loop — no-code, markdown-only. Gives any repository one committed memory/ that every AI agent and every contributor shares, with deterministic decay, supersession and a human-gated intent trace. Joined the family September 10, 2026. Accenture/mercury-go accenture.github.io/mercury-go
mercury-composable The Java engine — composable, event-driven applications in three layers: Platform Core, Event Script, Active Knowledge Graph. Accenture/mercury-composable accenture.github.io/mercury-composable
mercury The Rust engine — a twin implementation of mercury-composable: same three layers, same flow YAML (flow files port unchanged). Accenture/mercury accenture.github.io/mercury
mercury-python Python language pack — write composable functions in Python and call them from flows and knowledge graphs. Accenture/mercury-python accenture.github.io/mercury-python
mercury-nodejs Node.js language pack — write composable functions in Node.js and call them from flows and knowledge graphs. Accenture/mercury-nodejs accenture.github.io/mercury-nodejs

How the pieces fit

  • The engines (mercury-composable, mercury) run the application: functions addressed by route name exchange immutable events, YAML flows sequence them, and a knowledge graph provides the semantic layer.
  • The language packs (mercury-python, mercury-nodejs) extend the engines so functions can be written in Python or Node.js and called from the same flows and graphs.
  • agent-memory carries the project's memory — Vision, Blueprint, continuity, session records, architecture decisions — so the humans and AI agents building on the engines orient on purpose and state, not only on API shape. Every Mercury repository is itself AI-enabled with agent-memory: the family uses its own memory layer.

Using the brand

  • Short form: the Mercury family — advanced software foundations for human–AI collaboration.
  • Member sentence (for a member repository's README or site footer):

Part of the Mercury family — Accenture's open-source foundations for human–AI collaboration: agent-memory · mercury-composable · mercury · mercury-python · mercury-nodejs.

All members are published by Accenture under the Apache License 2.0.