Skip to content

Getting Started

SuperQode is the open-source, terminal-first Agent Engineering framework for your code factory, with first-class support for local and open models. Start by using it as a normal coding agent: open a repository, connect the agent or model you already use, and ask for coding work. You can switch the active harness during the session without giving up the familiar coding-agent workflow.

Harness authoring, evaluation, and optimization come later. They let you turn a useful interactive setup into a repository-owned, measured, and improvable run contract.

The Developer Journey

Follow the documentation in this order:

Stage What you do Start here
1. Use Install SuperQode and use it like another terminal coding agent Installation and Your First Session
2. Choose Connect any supported coding agent, model provider, local model, or optional harness Connection Methods and Vendors
3. Switch Change harnesses during a session while preserving or deliberately forking context Your First Session: Choose Or Switch A Harness
4. Build Create a repository-owned HarnessSpec when the default or existing harnesses are not enough Bring Your Own Harness
5. Evaluate Measure the harness against repeatable tasks and held-out checks Running, Measuring, and Optimizing a Harness
6. Optimize Use MetaHarness or GEPA Omni only after the evaluation contract is meaningful Optimization Story
7. Promote Stage, canary, activate, or roll back a proven harness Harness Promotion

Quick Navigation

  • Installation


    Install SuperQode and verify the command works.

    Install now

  • Use It As A Coding Agent


    Connect an agent or model, ask for coding work, and switch harnesses during the session.

    First session

  • Quick Start


    See the complete install, interactive, headless, harness, and inspection path in one guide.

    Quick start

  • Configuration


    Configure project defaults, providers, ACP agents, local models, and MCP servers.

    Configure

  • Harness Examples


    Start from ready-to-run HarnessSpec examples.

    Examples

  • Troubleshooting


    Diagnose install, provider, runtime, harness, sandbox, MCP, and session issues.

    Troubleshooting


Prerequisites

Requirement Version Notes
Python 3.12+ Required for the Python package
One-line installer recommended Installs uv when needed, then verifies an isolated SuperQode tool install
uv direct uv tool install superqode gives the same isolated tool install
Git 2.25+ Recommended for repository work and reviewing changes

Optional tools depend on your workflow:

Tool Use
Node.js and npm ACP agents, MCP servers, JavaScript tooling
Ollama, LM Studio, MLX, vLLM, SGLang, or DS4 Local model workflows
Docker or optional sandbox SDKs Container or remote sandbox profiles

1. Install

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

Installs uv when it is missing, then installs SuperQode in an isolated environment. No sudo required.

uv tool install superqode
superqode --version

Use this when you already have uv and prefer to manage the install yourself.

uvx superqode --version
git clone https://github.com/SuperagenticAI/superqode.git
cd superqode
uv sync --extra dev --extra docs
uv run superqode --version

Every method except source installs the latest SuperQode release from PyPI.


2. Start The TUI

cd /path/to/your/project
superqode

Connect a model or agent:

:connect

Direct examples:

:connect byok openai <openai-model>
:connect local ollama qwen3:8b
:connect acp opencode

Check state:

:status

Then ask a small first task:

Summarize this repository and suggest the smallest safe improvement.

3. Choose A Connection Path

Path Use when Setup
ACP You want an external coding agent with its own tool loop superqode agents list and superqode agents doctor <agent>
BYOK You want hosted providers with your own API keys Set OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY, or another provider key
Local You want private or offline inference Start Ollama, LM Studio, MLX, vLLM, SGLang, DS4, or another local server
SDK You want a supported vendor runtime or vendor-native agent behavior :connect codex, :connect antigravity, or an explicit API-key runtime in the TUI

See Connection Methods and Vendors for the supported connection methods, product profiles, providers, local engines, and ACP agents.

Provider diagnostics:

superqode providers doctor
superqode providers guide openai
superqode providers recommend coding

4. Create A Harness

A HarnessSpec makes a run repeatable. It controls runtime, model policy, tools, sandbox, approvals, checks, hooks, events, workflow, and output.

superqode harness init my-coder --template coding --output harness.yaml
superqode harness doctor --spec harness.yaml
superqode harness run --spec harness.yaml --prompt "summarize this repository"

Load it in the TUI:

:harness harness.yaml
:harness status

Built-in templates:

superqode harness list-templates

Common templates:

Template Purpose
coding Repository coding with file, search, edit, shell, todo, checks, and approvals
no-tool Model-only reasoning without repository or shell tools
gemma4-coding Gemma4 local coding starting point
gemma4-no-tool Gemma4 model-only reasoning
ds4-coding DS4 local coding starting point
ds4-fast-local Lower-latency DS4 local iteration

5. Run Headless

Use headless mode for scripts and one-off terminal tasks:

superqode --print "summarize this repository"
superqode --mode json --print "summarize this repository"
superqode --profile plan --print "plan the safest fix for the failing test"

Use a harness for repeatable headless behavior:

superqode harness run --spec harness.yaml --prompt "make the smallest safe fix and run the narrowest useful test"

6. Inspect Sessions And Runs

Sessions:

superqode sessions list
superqode sessions tree
superqode sessions show <session-id>
superqode sessions export <session-id> --format markdown --output session.md

Harness runs:

superqode harness runs
superqode harness events <run-id>
superqode harness graph <run-id>
superqode harness evidence <run-id>

Portable share artifacts:

superqode share create <session-id>
superqode share import <artifact.superqode-share.json> --session-id imported

7. Common CLI Commands

Command Purpose
superqode Launch the TUI
superqode --print "..." Run one headless task
superqode doctor Check core environment health
superqode config init Create superqode.yaml
superqode providers doctor Check provider setup
superqode agents list List known ACP agents
superqode runtime list List runtime backends
superqode harness init ... Create a HarnessSpec
superqode harness doctor --spec harness.yaml Preflight a harness
superqode harness run --spec harness.yaml --prompt "..." Run a harness task
superqode memory status Check memory providers
superqode sandbox doctor Check sandbox providers
superqode trust doctor Audit local project trust inputs

Which Workflow Should I Use?

Use the TUI when:

  • exploring a repository
  • making interactive changes
  • reviewing tool calls and approvals
  • switching providers or runtimes during a session
  • exporting or sharing a conversation

Use headless CLI when:

  • scripting a single task
  • running in automation
  • collecting JSON output
  • running repeatable HarnessSpec tasks
  • inspecting persisted run events

Next Steps

  1. Quick Start
  2. Your First Session
  3. Configuration
  4. Inside the Agent Loop, to understand what the engine is doing for you
  5. Harness System
  6. Connection Methods and Vendors
  7. CLI Reference