Skip to content
@x-mesh

x-mesh

Practical tools for building, running, and debugging AI agents.

x-mesh

Practical tools for building, running, and debugging AI agents.

open issues closed issues open PRs merged PRs

Commits on default branches and merged pull requests per month, across every x-mesh repository including private ones. Counts are in activity.json.

x-mesh is a collection of independent tools for agent development and production operations. Each tool works on its own, and integrations are explicit and optional.

The projects follow the same operating rules: inspect before changing anything, keep changes reversible, and report measurements instead of guesses.


The layers

Five independent tool groups: agent environments, development workflow, session context, version control, and production diagnostics.

Rendered by card from card.json. The repository count is generated rather than hard-coded.

Layer Purpose
Environment term-mesh Runs multiple agents in isolated worktrees, locally or over SSH.
Harness xm Plans small repository changes from code evidence and reviews the result with multiple models before merge.
Context mem-mesh · output-mesh mem-mesh stores decisions and resumable work state. output-mesh records agent artifacts and the sessions that produced them. Both are optional, read-only integrations configured per agent.
Version control gk Adds reflog-based undo, snapshot restore, and repository policies.
Operations aic · edc · httprove · dbops · x-backup Diagnoses shell, system, network, HTTP, database, and backup problems. Diagnostic actions are read-only by default. Commands that write require an explicit flag and offer a dry run.

The tools do not depend on one another. mem-mesh is an MCP server, edc and httprove are standalone CLIs, and xm works without term-mesh.

Integrations use explicit contracts. xm and term-mesh share integration specs, and gk finish --gate sends a completed worktree to xm for review. mem-mesh is configured per agent and works with any supported environment. Without it, term-mesh still runs but starts each session without saved context.

Design rules

term-mesh leaves the context limit unknown when it does not recognize a model:

A percentage computed against a guessed denominator looks exactly like a measured one.

A guessed limit would make an estimate look measured. The same rule applies elsewhere: diagnostics do not make changes by default, write operations are explicit, and reported figures come from collected data.

Also here
  • space-mesh: macOS disk space analyzer. SwiftUI interface, Rust scanning core.
  • headroom: Browser tool for topology authoring and infrastructure constraint analysis. Shows which capacity axis saturates first, in the browser, calculations kept local.
  • clear-korean: Instruction set that makes AI answer in short, precise Korean.
  • homebrew-tap: brew tap x-mesh/tap

Pinned Loading

  1. xm xm Public

    Plugin toolkit for Claude Code — grounds plans in repository evidence, gates the result before it merges

    JavaScript

  2. mem-mesh mem-mesh Public

    Persistent memory for AI agents — hybrid vector + FTS5 search

    Python

  3. edc edc Public

    everyday carry for SE and SRE — a read-only network and system diagnostic CLI for Linux and macOS

    Go

  4. aic aic Public

    Terminal LLM assistant — shell-error analysis and an SRE agent that runs bounded, read-only diagnostics

    Rust

  5. httprove httprove Public

    HTTP(S) diagnostics for SREs — per-phase latency waterfall, TLS/chain inspection, health verdict, per-backend fanout, continuous probing, TUI

    Rust

  6. term-mesh term-mesh Public

    AI agent control plane for macOS — parallel agents, each in its own sandboxed worktree, local or over SSH

    Swift

Repositories

Showing 10 of 17 repositories

Top languages

Loading…

Most used topics

Loading…