ox, built by Sageox, is a context-engineering tool that acts as a shared memory for human-AI development teams, addressing context drift between sessions. The tool captures conversations, Slack threads, and coding sessions into a persistent Ledger and primes AI assistants at session start. It includes command-line utilities, automated distillation into Knowledge Bubbles, and a Murmur coordination channel. Target users are AI-native engineering teams seeking consistent architectural intent across multi-session workflows.

What tasks can you actually use it for?

The tool targets continuous team memory across multi-session development, operating as an MCP server and a CLI utility. It captures decisions and coding context from meetings, chat threads, and live coding into a persistent, queryable Ledger, then injects that material into agent sessions with the Agent Prime command. Use cases include preserving architectural intent across handoffs, providing context to coding assistants, and keeping multiple agents aligned during long-lived engineering projects.

How reliable is the shared memory and context delivery?

Ox stores captured material in a persistent, queryable record and distills conversations into Knowledge Bubbles, which are automatically loaded for agents. Those two mechanisms are designed to reduce context drift by preserving prior decisions and conventions. Murmur provides real-time coordination signals so agents can avoid duplicating work. Reliability depends on how comprehensively the Ledger captures relevant artifacts and how consistently teams use the system.

What inputs and integrations does it require?

The tool integrates with MCP-compatible hosts and popular coding assistants, and it is agent-agnostic while being optimized for Claude Code. Supported clients listed include Cursor, Windsurf, Aider, and Codex CLI. It runs as a CLI on macOS, Linux, and FreeBSD and can be installed via a shell script or built from source using Go. Those platform and client requirements define where the tool adds value.

Is it practical to adopt in an existing developer workflow?

Adoption favors teams already using MCP-compatible agents and comfortable with command-line tooling. The open-source MIT license and a build-from-source path let teams inspect and modify behavior, but initial setup requires CLI installs and integration with existing chat and coding channels. Teams relying on GUI-only assistants or without MCP hosts face additional integration work before the Ledger and Murmur signals deliver coordinated benefits.

A targeted option for AI-native engineering teams with MCP workflows

Ox is a practical choice for engineering teams that run MCP-compatible agents and need continuous cross-agent memory to preserve architectural intent. It depends on an MCP-based workflow and command-line integration to provide value, so teams without those elements should expect setup overhead. For groups aiming to reduce session fragmentation across multiple AI assistants, the tool provides a focused way to keep decisions discoverable and actionable.