TERMINAL FIRST · ANY AGENT · ANY MODEL · LOCAL OR CLOUD · OPEN SOURCE superqode.dev

THE HARNESS HUB FOR CODING AGENTS

HARNESSLAYERfor coding agents

The Harness Hub where coding harnesses are discovered, built, inspected, run, evaluated, and optimized through one rich, browsable terminal experience.

One portable HarnessSpec controls runtime, model policy, tools, memory, search, sandbox, approvals, workflow, and evidence. Keep the agent you trust. Own the loop around it.

superqode - install

# one line, no sudo, isolated env

$ curl -fsSL https://super-agentic.ai/superqode.sh | sh

cd your-project && superqode # alias sq

Python 3.12+ · macOS 12+ · Linux · WSL2Other options →

Meet the A2A agent Explore the Hub 104 harnesses

superqode.dev

104

harnesses

open source

spec

Harness Hub

Browse every route in the terminal

:hub

A2A Agent

Any client can ask which harness to use

A2A

One evidence graph

Same events across every runtime

JSON

Terminal first. Evidence is portable.

APACHE 2.0

License

PYTHON 3.12+

Runtime

ONE SPEC

Portable YAML

KEEP THE AGENTS YOU ALREADY PAY FOR

ALL CONNECTIONS →

Codex

Claude

Copilot

Grok

Antigravity

Kimi

Qwen

GLM

Gemma

DeepSeek+ ACP registry · 40 BYOK providers · local engines

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 key.

NEW · AGENT2AGENT

SuperQode A2A Agent

SuperQode publishes an Agent Card and answers Agent2Agent requests. Any A2A client can discover it, ask which coding agents and harnesses to consider, and get an answer without sending a repository anywhere.

Read the A2A guide →

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.

Signed API keys

When you need to gate access, keys carry a customer, tier and expiry, and verify with a signature check and clock comparison. The key holds its own claims, so verification survives a redeploy.

Safe by default over the wire

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

DISCOVERY

Published Agent Card

Any A2A client can discover SuperQode, ask which coding harnesses to consider, and get a ranked shortlist without sending a repository.

A2A guide →

super-agentic.ai/.well-known/agent-card.json

OPTIMIZATION

Measure, then improve with evidence.

Optimization never edits the live spec. Candidates stay reviewable artifacts gated by repository tasks. You promote explicitly.

GEPA Omni · HarnessBench · SkillOpt

GEPA Omni

Three optimizers, one promotion path

Generate candidates, mine failures, and select on the Pareto frontier with a sealed held out gate. Stages its selected spec separately.

HarnessBench

Benchmark on your repo

Repeatable benchmark packs with manifest, scorecard, and drift checks. Same spec runs in CI.

SkillOpt

Optimize skills safely

Improve reusable skills and recipes under the same gates, with guardrails on mutation surfaces.

Run, Measure, Optimize Optimization story Harness promotion

THE PROBLEM

Coding agents alone do not create a dependable code factory.

Every agent keeps its own context, tools, permissions, sessions, and evidence. SuperQode provides one engineering lifecycle across them.

Agent adoption is fragmented

Teams can choose specialized coding agents and models for different work. Each product keeps separate sessions, context, tools, permissions, and execution evidence.

The harness is rarely an artifact

Vendor tuned loops provide useful defaults. Teams still need to inspect, version, evaluate, and adapt the loop for their own repositories.

Output is not verified delivery

Agents can plan, edit, test, and review. Orchestration alone does not provide isolation, recovery, policy enforcement, evidence, and explicit delivery.

Code factoryThe organization owned system of coding agents, harnesses, models, repositories, policies, evaluations, and delivery workflows that turns intent into verified code changes.Factory guide →

HARNESS OWNERSHIP

Own the loop,
not just the model.

SuperQode keeps agent configuration, context, memory, tools, policies, workflows, and evaluation criteria in a versioned HarnessSpec maintained with your code.

- Portable across Codex, Claude, Copilot, and local models

- Inspectable policy for tools, sandbox, and approvals

- Same evidence graph regardless of runtime

Repository owned specification

Context, memory, tools, workflows, policies, and evaluation criteria in a versioned spec.

Runtime independence

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

Controlled improvement

Evaluate candidates against repository tasks and promote only after gates pass.

Harness independenceKeep the product you rely on and benchmark it against a controlled alternative on the same repository.Build a harness →

OPEN
55 harnesses

CLOSED
vendor plans

LOCAL
airplane mode

LIFECYCLE

Build, run, evaluate, govern, optimize.

Every stage emits the same event graph so results are comparable regardless of runtime.

SPEC

Choose coding, no tool, or custom behavior

MODEL

Apply policy, local hints, fallbacks

RUNTIME

builtin, Codex SDK, ACP, ADK, PydanticAI

TOOLS

File, search, shell, MCP, or no tools

SESSION

Persist, compact, fork, handoff, resume

OUTPUT

Text, diffs, typed data, evidence

CAPABILITIES

Everything needed to run the factory

Interactive TUI

Switch models, pick harnesses, inspect sessions without leaving the terminal.

HarnessSpec

Portable YAML contract for tools, permissions, model policy, memory, and workflows.

WorkOrders

Dependency aware tasks with isolated workers, recovery, and delivery gates.

Agent connections

Local runtimes, 40 BYOK providers, ACP agents, and native SDKs.

Remote control

Telegram, Slack, Discord, or browser TUI with allowlists and approvals.

Sandboxed execution

Local OS, Docker, Podman, or hosted E2B, Daytona, Modal, Vercel.

Headless and CI

Same specs, policies, and evidence noninteractively in pipelines.

Evaluate and optimize

Score behavior, compare candidates, then promote or roll back.

Safety and policy

Approvals, sandbox, trust checks, and plugin policy explicit and inspectable.

Local and open models

Ollama, MLX, LM Studio, vLLM, SGLang, TGI, llama.cpp, offline.

Sessions

Persistent sessions with lineage, forks, handoffs, and export.

Recursive models

Native RLM, RLM Code, Prime Agent, and recursive tools for large artifacts.

HARNESS HUB

Find the right harness for the job.

Native harnesses, coding agents, ACP routes, optional runtimes, and model presets with runtime, setup, provenance, continuity, and safety shown up front.

How the Hub works

Hub104 harnesses · Filter by Ready · Open source · Your project harnessesBrowse in the terminal

HARNESS

RUNTIME · CATEGORY

STATUS

Core

Default loop with policy stack

builtin · SuperQode harnesses

Ready

Workbench

Wider tool surface, same approvals

builtin · SuperQode harnesses

Ready

RLM

Recursive Python harness

builtin · Python · Docker / Monty

Ready

PiPy

Event first Python harness

builtin · pipy

Ready

Codex

Codex subscription plan

codex-sdk · Coding agents

Managed

OpenCode

OpenCode via ACP

acp · ACP agents

ACP

Inspect details inside the terminal. No need to leave the session.Browse all 104 →

RECURSIVE · PYTHON

Native RLM harness

One executable tool and a persistent Python environment. Build context by writing Python instead of calling separate search and edit tools. Host, Docker, or Monty.

CODE FACTORY

WorkOrders that finish

Durable, dependency aware tasks with isolated worktrees, bounded workers, crash recovery, and explicit human approval.

LOCAL AND OPEN

Airplane Mode and local search

Auto context management, bounded reads, ripgrep plus semantic search, and strict offline mode.

MAINTAINED HARNESS PRESETS

Harness system docs →

SuperQode

Kimi

GLM

Qwen

DS4

Gemma

MiniMaxEvery install ships these model family presets

START HERE

Install,
open a repo,
start.

Everything after this runs from the TUI or the CLI. The short alias sq runs every superqode command.

Already have uv? uv tool install superqode or uvx superqode

Remove any time: uv tool uninstall superqode

Omnigent agent specifications can be imported as portable HarnessSpecs. Omnigent docs →

Inside the terminal

Browse the Hub, switch harnesses mid session, fork attempts, and export portable session evidence. Headless and CI use the same spec and governance.

Headless

Same spec in pipelines

Sessions

Durable, forkable

Branches

:harness --fork

FULL REFERENCE

Every command and option is documented

Installation, provider setup, HarnessSpec schema, evaluation workflows, and the complete CLI. Same evidence in terminal and CI.

Browse the documentation View on GitHub

A2A HARNESS FIT · DESIGN PARTNERS

Harness recommendation with evidence.

New harnesses land every week. Choosing a coding agent is a spending decision that almost nobody has evidence for. The A2A agent narrows the field today. Partners help shape the measurement that decides it.

Fit Check

Free, available now

Browse the Hub, explain a spec, check drift, and run bundled evaluation packs on your machine.

Fit Report

One off

Benchmark a chosen set of harnesses against your repository. Scorecard with Pareto frontier, reproducible manifest, and written recommendation.

Fit Watch

Recurring

Re runs when a new model or harness ships, or when a spec drifts. Tells you when the winner changes.

Fit Private

Enterprise

Same container in your own cloud. Your code stays in your network, private Hub entries for internal harnesses.

Fit Check ships today. The other three are what the program is for. Partners help decide what they measure before anything is priced.

Contact design partners partner@super-agentic.ai

FAQ

Frequently asked questions

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

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 →

The harness layer, versioned with your code.

Build harnesses you own, or connect the agents you already use, then evaluate and improve how they work across repositories.

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

Install SuperQode Read the documentation

HARNESSLAYERfor coding agents

SuperQode A2A Agent

Published Agent Card

Measure, then improve with evidence.

Coding agents alone do not create a dependable code factory.

Own the loop,not just the model.

Repository owned specification

Runtime independence

Controlled improvement

Build, run, evaluate, govern, optimize.

Everything needed to run the factory

Find the right harness for the job.

Native RLM harness

WorkOrders that finish

Airplane Mode and local search

Install,open a repo,start.

Inside the terminal

Every command and option is documented

Harness recommendation with evidence.

Frequently asked questions

The harness layer, versioned with your code.

Own the loop,
not just the model.

Install,
open a repo,
start.