I/O Mesh

Open source · Resource

Open-source projects

Five MIT-licensed repositories on GitHub under iome-sh. Each card below covers what the project is, how to install it, and where it fits in a local or mesh-connected agent stack.

All five projects are public MIT source. They are designed to work together: Go and Python clients for mesh I/O, TUI/agent harness for operators, local-edge Memory kernel for recall primitives, and iomesh-memory-mcp as a lean MCP host for that kernel.

Clone, build, and inspect the code on GitHub. SemVer 1.0 TUI plus MIT edge repos; each repo’s README and CI document the supported contracts.

Browse all repos on GitHub

Edge-first path

From signup to local analyze

Residual-honest first run: portal for workspace keys and connector installs, open-source TUI + local Memory MCP on the edge — not freemium hosted Memory GA.

  1. Step 1

    Portal

    Sign up

    Create a workspace in the portal (Base from $132/mo when you need the mesh). Self-serve — no waitlist as the primary path.

  2. Step 2

    iomesh-tui

    Download TUI

    Install MIT iomesh-tui from GitHub (v1.0 public harness) — multi-provider agent TUI/CLI with optional mesh hooks, MCP, and residual-honest setup continuum. See README · make build.

  3. Step 3

    Portal + TUI

    Keys

    Portal issues mesh/API keys. Configure LLM provider keys in the TUI yourself — models are not platform-bundled by default.

  4. Step 4

    iomesh-tui + Portal

    Setup continuum

    Run setup init → preflight → portal next steps (session path stays in-TUI; cold path restarts then preflight). Browser portal HITL for connector installs; agent MCP cannot write installs; catalog ≠ Connected.

  5. Step 5

    Portal

    Integrations next step (HITL)

    After integrations list/plan/status, OAuth and install write stay in the browser portal. Catalog ≠ Connected; template plan ≠ install apply; package wire ≠ Connected. Events land on the mesh under tenancy when live.

  6. Step 6

    iomesh-memory-mcp

    Local memory MCP

    Install github.com/iome-sh/memory + iomesh-memory-mcp for local recall. Optional mesh pull is separate.

  7. Step 7

    TUI + MCP

    Analyze

    TUI + your model analyze on local Memory tools and optional mesh context — policy-gated MCP, not a chat seat trial.

How the public projects fit together

Use iomesh-client-sdk-go or iomesh-client-sdk-python for mesh publish/consume, iomesh-tui as a full agent harness, github.com/iome-sh/memory as the local-edge Memory kernel library, and iomesh-memory-mcp as the lean MCP host for that kernel. Together they cover client I/O, interactive agents, and local-primary memory tools.

The public repositories are the inspectable edge: Go and Python SDKs, iomesh-tui, the memory kernel, and iomesh-memory-mcp. Clone, install, and run them from GitHub.

Context horizons agents can work across

Public clients and the TUI work with live streams, short rolling history, and longer patterns under the policies of the environment you configure.

Real-time

Live pulse rates and ordered stream state. Client SDK publish/pull and TUI mesh hooks subscribe when a mesh endpoint is configured.

Short-term

Hours to days of stream history and rolling operational context agents can cite (recent tickets, deploys, stages).

Long-term

Longer analytical or institutional patterns via the memory kernel and iomesh-memory-mcp on your machine.

The full organizational heartbeat — three connected signals

One fabric, three signals: live operations, knowledge, and analytics — department-scoped, policy-gated.

Live operations

Tickets, incidents, deploys, and customer updates, organized by team. The primary heartbeat a reviewer can see after a source, a stream, and a signed event.

Knowledge

Docs, runbooks, and memory of past heartbeats. What previous activity meant. Reviewers can open this layer in the product; do not chip it as a readiness label.

Analytics

Patterns and trends across heartbeats so agents see beyond the moment. Same rule: describe the layer, do not chip a status.

Project inventory

Public MIT repositories: iomesh-client-sdk-go · iomesh-client-sdk-python · iomesh-tui · iome-sh/memory · iomesh-memory-mcp. See each repository README for install, configuration, and release notes.

Five public MIT repositories under github.com/iome-sh

Local-edge Memory: go get github.com/iome-sh/memory · go install github.com/iome-sh/iomesh-memory-mcp/cmd/iomesh-memory-mcp@main

Local-first agent & memory FAQ

Short answers for operators evaluating the open-source edge before (or without) a mesh subscription.

Is I/O Mesh open source?

The public developer stack is MIT open source: Go client SDK, early-Beta Python client SDK, iomesh-tui agent harness, local Memory kernel, and iomesh-memory-mcp. Portal keys, connector installs, and the governed ops mesh ship with paid Base Plan — separate from the free local path, not freemium hosted Memory GA.

How do demos work?

Demo the edge yourself: install iomesh-tui as the local agent harness and optionally run iomesh-memory-mcp for local-primary memory. That path needs no cloud subscription and no hosted demo environment. Sign up for Base Plan when you want portal keys, connector installs (HITL), and the governed ops mesh.

What is iomesh-tui?

iomesh-tui is an MIT-licensed agent harness on GitHub — a multi-provider coding agent TUI/CLI with optional mesh hooks and MCP.

Local memory vs cloud memory?

The memory kernel and iomesh-memory-mcp run on your machine. Mesh Base is a separate paid product.

Does it require IOMesh Cloud?

No. You can run iomesh-tui and local memory tools without a cloud subscription. Sign up for Base Plan when you need portal keys, connector installs (HITL), and the governed ops mesh.

How do I try local memory?

Install github.com/iome-sh/memory and iomesh-memory-mcp from GitHub, then point your agent or TUI at the local MCP host. See each repository README for install and configuration steps.

iomesh-client-sdk-go

MITGo

MIT · Official Go client

Public · official client

https://github.com/iome-sh/iomesh-client-sdk-go

Official Go client for publishing to and consuming from I/O Mesh brokers and connectors. Covers HTTP publish/pull, streams, KV, memory hooks, partner webhook HMAC helpers, and a Kafka Produce subset.

  • Install: go get github.com/iome-sh/iomesh-client-sdk-go
  • Wire headers and IOMESH_* environment conventions documented in README
  • CI-tested surface for agent and service integrations
  • Module path: github.com/iome-sh/iomesh-client-sdk-go

go get github.com/iome-sh/iomesh-client-sdk-go

Why we made it public: Use this when you want a small, inspectable Go library for mesh publish/consume without vendoring a monorepo.

Open full repo → https://github.com/iome-sh/iomesh-client-sdk-go

iomesh-client-sdk-python

MITPython

MIT · Official Python client

Public

https://github.com/iome-sh/iomesh-client-sdk-python

Official Python client for mesh publish and consume. See the repository README for install and the current API surface.

  • Install and usage: see repository README
  • HTTP plane · publish/pull · KV · catalog/policy helpers
  • MIT · public under github.com/iome-sh/iomesh-client-sdk-python
  • MIT · public under github.com/iome-sh/iomesh-client-sdk-python

See README · Python client

Why we made it public: Use this when you want mesh publish/consume from Python without wrapping the Go SDK.

Open full repo → https://github.com/iome-sh/iomesh-client-sdk-python

iomesh-tui

MITGo

MIT · Agent harness & TUI · v1.0

Public foundation · SemVer 1.0 · v1.0 packaging

https://github.com/iome-sh/iomesh-tui

Multi-provider coding-agent TUI/CLI with optional mesh hooks, MCP, skills, and workspace tools.

  • Build: see README · make build · pin v1.0+ releases from GitHub
  • Setup: init, preflight, then browser portal for OAuth and installs
  • Optional model pins (gemini-*, vertex-gemini-*, and others)
  • Subagents, MCP, skills, ACP, workspace tools in one harness

See README · make build · pin models via iomesh-tui (v1.0+)

Why we made it public: A runnable, model-agnostic agent harness you can audit end-to-end.

Open full repo → https://github.com/iome-sh/iomesh-tui

memory

MITGo

MIT · Memory kernel

Public · local-edge kernel

https://github.com/iome-sh/memory

Local-edge Memory kernel (Go module github.com/iome-sh/memory). Library of memory primitives for local-primary agent recall — store, search, and temporal options for edge workloads.

  • Install: go get github.com/iome-sh/memory
  • Library module — no binary required for basic use
  • Local-first recall primitives
  • RELEASING.md documents semver tags for module consumers

go get github.com/iome-sh/memory

Why we made it public: Inspect and extend the kernel that local Memory tools build on — pure library surface for edge recall.

Open full repo → https://github.com/iome-sh/memory

iomesh-memory-mcp

MITGo

MIT · Edge MCP host

Public · lean edge host

https://github.com/iome-sh/iomesh-memory-mcp

Lean edge MCP host binary that exposes local-primary Memory tools over MCP. Product name is iomesh-memory-mcp. Ships multi-platform releases with SBOM and keyless cosign on checksums.

  • Install: go install github.com/iome-sh/iomesh-memory-mcp/cmd/iomesh-memory-mcp@main
  • Depends on public github.com/iome-sh/memory (no GOPRIVATE required)
  • Local-first MCP host for the memory kernel
  • GoReleaser + SBOM + keyless cosign on version tags (see RELEASING.md)

go install github.com/iome-sh/iomesh-memory-mcp/cmd/iomesh-memory-mcp@main

Why we made it public: Run Memory as an MCP server on the edge with a small, installable host you can read and rebuild.

Open full repo → https://github.com/iome-sh/iomesh-memory-mcp

Optional model pins in the TUI

iomesh-tui can pin models your team already uses (including Gemini-class models when configured). Agent LLM spend is opt-in configuration — not required to build or run the open-source harness.

What you get from these projects

  • Inspectable mesh clients — publish/consume contracts in iomesh-client-sdk-go and early-Beta iomesh-client-sdk-python.
  • Runnable agent harness — multi-provider TUI with MCP, skills, and optional mesh hooks.
  • Local-edge Memory kernel — github.com/iome-sh/memory for library-level recall primitives.
  • Edge MCP host — iomesh-memory-mcp turns the kernel into installable MCP tools.
  • Standard OSS workflow — public GitHub, MIT license, CI, and documented install paths.

Issues and PRs welcome on every MIT repo (Go SDK, Python SDK, TUI, memory kernel, iomesh-memory-mcp). Prefer small, CI-green changes with clear README updates.

Built with

  • Go · Python · MIT
  • MCP-ready edge host
  • Local-edge Memory kernel
  • Optional agent LLM pins

Five public projects, one clear stack

Start with the Go SDK (or Python SDK) when you need mesh I/O from services or agents. Use iomesh-tui when you want a full agent harness. Use the memory kernel when you need local recall as a library. Use iomesh-memory-mcp when agents should call that memory over MCP.

Typical path: sign up, install iomesh-tui, create keys, then use the browser portal for connector installs. LLM weights come from your providers.

Every repository is MIT-licensed and public under github.com/iome-sh. Read the README in each repo for install, configuration, and contribution guidelines.

Next steps

Clone the repos, install iomesh-tui and iomesh-memory-mcp, then use the browser portal for connector installs. See Pricing when you are ready to connect the mesh.

Create a workspaceSee pricing
  • MIT license
  • Public GitHub (iome-sh)
  • Five open repositories
  • Go and Python clients
  • Optional agent model pins

MIT · iomesh-client-sdk-go · iomesh-client-sdk-python · iomesh-tui · memory · iomesh-memory-mcp