SuperQode is the open-source, terminal-first harness layer for coding agents. Build repository-owned HarnessSpecs, run SuperQode's native harnesses, or connect established open-source and proprietary agents through native integrations, SDK runtimes, RPC, and ACP.

What is a coding-agent harness?

A coding-agent harness is the runtime and control system around a model. It manages the agent loop, codebase context, memory, tools, shell and sandbox access, workflows, permissions, evaluations, and recovery required to produce verified code changes. SuperQode defines that harness as code through a repository-owned HarnessSpec.

Do I have to replace the coding agents my team already uses?

No. Keep Codex, Claude, Copilot, Grok, and the rest as the agents you write code with. SuperQode connects them through native integrations, SDK runtimes, and ACP, then versions and scores the harness on your repository.

Does SuperQode work with local and open models?

Yes. SuperQode is optimized for local and open models including Ollama, MLX, LM Studio, vLLM, SGLang, TGI, DS4, and llama.cpp, and Airplane Mode supports strict offline work. It also supports BYOK cloud models and subscription runtimes through the same harness layer.

How do I install SuperQode?

Run the one-line installer: curl -fsSL https://superqode.dev/install.sh | sh. It uses an existing uv installation when one is available and otherwise runs the official uv installer. You can also install with "uv tool install --upgrade --force superqode", or run it without installing using "uvx superqode". SuperQode requires Python 3.12 or 3.13 and runs on macOS 12+, Linux, and Windows 10+ through WSL2. The short alias "sq" runs every superqode command.

Do you work with design partners?

Yes. Organizations evaluating SuperQode for team or multi-agent workflows can contact partner@super-agentic.ai.

Harness layer · Open source

SuperQode

Live on Product Hunt

The Harness Layer for Coding Agents

Find a coding harness, shape it to your repository, run it, and measure whether it actually did better. All from one terminal.

Keep the coding agent you already pay for. SuperQode is the layer around it.

install superqode

$curl -fsSL https://superqode.dev/install.sh | sh

Python 3.12+. uv, uvx, and other install options

Go to the official site

Install, the Harness Hub, the A2A agent and the docs all live on superqode.dev.

Harness Hub A2A Agent GitHub Docs

Agent Card

a2a.superqode.dev

Live

"name": "SuperQode"

"protocolVersion": "0.3"

"preferredTransport": "JSONRPC"

"url": "https://a2a.superqode.dev"

Skills

Harness Shortlist

Ranked third party candidates from the Harness Hub.

Interfaces

A2A 1.0A2A 0.3HTTP+JSON

100% MUST on the official TCK.

Call the A2A agentlive

Any A2A client can ask which coding agents to consider.

Demo

See it run

One session: connect an agent, browse the Harness Hub, and call the A2A agent from the same terminal.

Watch on YouTube

Connect the agents you already use

Codex

Claude

Copilot

Antigravity

Grok

Prime Agent

Muse Code

Kimi

GLM

Plus local models and BYOK providers. Providers docs

Four ways to connect

Subscriptions

Agent Client Protocol

Any ACP agent, driven from the same terminal.

Open harnesses

OSI-licensed harnesses on your own key or a local model.

Closed harnesses

Proprietary harnesses on that vendor's key.

Open protocols

ACPexternal coding agents

MCPtools and resources

A2AAgent Cards other agents can callNew

UHPharnesses on a remote serverNew

Harness ownership

One control plane

Configuration, context, tools, policies, and evaluation live in a HarnessSpec versioned with your code.

Spec in the repo

Maintain context, memory, tools, workflows, policies, and evaluation criteria in a versioned HarnessSpec.

Any agent

Use Codex, Claude, Copilot, Grok, OpenCode, local models, and compatible ACP agents with the same repository controls.

Then measure

Evaluate harness candidates against repository tasks and promote changes only after the required gates pass.

Capabilities

Harness capabilities

Connect the daily agent, specify the loop, and measure it on your repository.

Connect

Keep the agents you already pay for.

Interactive TUI

Switch models and harnesses in one terminal.

Agent connections

Subscriptions, ACP, BYOK, and vendor SDKs.

Local models

Ollama, MLX, LM Studio, vLLM, llama.cpp.

Chat and browser

Telegram, Slack, Discord, or a hosted TUI.

Specify

The contract around the agent.

HarnessSpec

YAML for tools, permissions, policy, memory.

Sandbox

Local OS, Docker, Podman, E2B, Daytona, Modal.

Safety and policy

Approvals, trust checks, plugin policy.

Sessions

Lineage, mid-session switch, forks, export.

Measure

Score the harness on your repository.

Evaluate

Compare candidates, then promote or roll back.

WorkOrders

Isolated tasks, recovery, evidence, gates.

Headless and CI

Same spec and evidence in a pipeline.

RLM routes

Host, Docker, Monty, RLM Code, Prime Agent.

Harness Hub

The Hub

Browse native harnesses, coding agents, ACP integrations, optional routes, and maintained presets. Each entry lists runtime, setup, provenance, continuity, and safety.

105 catalog entries

Explore Harness Hub :hub

Agent2Agent

A2A agent

Any A2A client can ask which coding agents to consider. SuperQode can also connect to an Agent Card from the TUI or CLI. Open cards need no token, and the agent passes every MUST in the official A2A Technology Compatibility Kit.

One card, three interfaces

The published Agent Card advertises JSON-RPC and HTTP+JSON across A2A 1.0 and 0.3. Gemini Enterprise, Microsoft Foundry, and Bedrock AgentCore all accept that same document.

A vendor-neutral shortlist skill

Ask which coding agents to consider and get ranked third-party candidates from the Harness Hub with licence and setup details. SuperQode harnesses are excluded from the ranking and disclosed separately.

Graded against the official TCK

Scored with the Technology Compatibility Kit the A2A project maintains under the Linux Foundation: every MUST and every MAY passes, for 96.4% overall.

Signed API keys

Open calls use keyword matching. Email hello@super-agentic.ai for a key; the request is then read by a model. Keys carry a customer, tier and expiry.

Shortlist only, over the wire

A remote bind serves the Hub shortlist, rate limited per caller and per day. Running harnesses remotely requires --expose-harness, a named spec, and a token.

serve

# localhost: the harness skill is served
superqode serve a2a --spec harness.yaml

# remote bind: shortlist only, unless you opt in
superqode serve a2a --host 0.0.0.0 --allow-remote

connect

# fetch any Agent Card from SuperQode
:connect a2a
superqode connect a2a --url https://a2a.superqode.dev --inspect
superqode connect a2a --url https://a2a.superqode.dev --conformance

Discovery

The public Agent Card lives at a well-known path and names the interfaces a client should call. From SuperQode, :connect a2a fetches that card, or any other origin, and lists what it advertises.

a2a.superqode.dev/.well-known/agent-card.json

The SuperQode A2A Agent Read the A2A guide

Start here

Install and connect

Then work from the TUI or CLI. The alias sq runs every superqode command.

terminal

curl -fsSL https://superqode.dev/install.sh | sh
cd /path/to/your/project
superqode
# inside the TUI: :hub

Connect local models, BYOK providers, ACP agents, and native vendor runtimes. See the Connections docs.

Omnigent agent specifications can be imported as portable HarnessSpecs. See the Omnigent compatibility docs.

Maintained harness presets

SuperQode

Kimi

GLM

Qwen

DS4

Gemma

MiniMax

Every install ships maintained presets for these model families. Harness system docs

Documentation

Reference

Everything from the one-line install to the HarnessSpec contract, connecting agents over ACP, MCP, UHP and A2A, and running HarnessBench against your own repository. Every command and flag is documented, along with the environment variables each one reads.

Browse the documentation View on GitHub

Partner program

Design partners

Limited seats for teams standardising coding-agent workflows. Partners help decide what gets measured on their own repositories.

A2A Harness Fit

Harness Recommendation and Optimization

The public A2A agent shortlists from the Hub. Partners shape the bench that ranks harnesses on a given repository.

Free

Open source, now

The full terminal and the optimizers, all 105 harnesses in the Hub, and the hosted A2A agent with no key.

One off

A single engagement

Benchmark candidates on your repository. Scorecard, reproducible manifest, written recommendation.

Recurring

Ongoing

Continuous spend measurement, re-runs when a model or spec changes, in our cloud or self-hosted in yours.

The free tier is available now. The other two are shaped with partners. Full details on superqode.dev

Contact design partners partner@super-agentic.ai

The harness layer

Build your own harnesses or connect open-source and proprietary agents, then evaluate, govern, and improve how they work across your repositories.

Open source · Terminal first · Any agent or model · Local or cloud

Get SuperQode at superqode.dev View on GitHub Read documentation

Design partners

FAQ

Questions

What SuperQode is, which agents it connects, and how to install.

SuperQode is built and maintained by Superagentic AI. We also work with teams directly on agent experience, engineering, optimization and delivery. See how we work with teams.

SuperQode

The Harness Layer for Coding Agents

See it run

One control plane

Spec in the repo

Any agent

Then measure

Harness capabilities

Connect

Interactive TUI

Agent connections

Local models

Chat and browser

Specify

HarnessSpec

Sandbox

Safety and policy

Sessions

Measure

Evaluate

WorkOrders

Headless and CI

RLM routes

The Hub

A2A agent

One card, three interfaces

A vendor-neutral shortlist skill

Graded against the official TCK

Signed API keys

Shortlist only, over the wire

Install and connect

Reference

Design partners

Harness Recommendation and Optimization

Free

One off

Recurring

The harness layer

Questions

What is SuperQode?

What is a coding-agent harness?

Do I have to replace the coding agents my team already uses?

Does SuperQode work with local and open models?

How do I install SuperQode?

Do you work with design partners?