AI builder agent for hackathons, startups, and open-source projects

CodeThon CLI

Plan the product, initialize the workspace, generate the roadmap, execute the build, debug the failures, profile the code, and prepare the launch from one OLED-dark terminal experience.

Install the CLI Explore builder workflow

npm install -g codethon-cli

For buildersOpen sourceProvider agnosticSlash-command REPLStreaming outputGuarded execution

ct interactive builder workspace

$ct

CODETHON CLI · OLED Workspace · AI READY

Project: CodeThon-CLI · Provider: NVIDIA · Model: deepseek-v4-flash

/execute <goal> Autonomous OLED workspace

/inspect [runId] Inspect execution journal

/replay [runId] Replay event timeline

/graph [dir] Repository graph view

/memory [query] Memory explorer

>/init

[ok] Workspace registered. Next: /plan

>/execute build auth and add tests

[Planner] Visible plan · [Scout] reading files · [QA] running checks

[receipt] completed · journal saved · next /inspect /replay /diff

Workflow

From idea to a working agent loop.

CodeThon is designed for builders who need momentum: connect a model, run /init to shape the workspace, use /planto design the path, then give /execute a concrete task. Diagnostics, profiling, recovery, and launch support stay in the same terminal.

/auth add

1. Connect

Choose the model stack.

Configure hosted or local models, validate access, and keep credentials outside the source tree.

/init

2. Initialize

Turn the idea into a workspace.

/plan

3. Plan

Get the roadmap before the code.

Stream a builder-friendly plan with milestones, architecture, data flow, risks, and the fastest path to a demo.

/execute <goal>

4. Execute

Let the agent work through the task.

Read files, apply edits, run guarded commands, repair errors, and keep the user informed while it works.

/doctor

5. Verify

Analyze, profile, recover, and ship.

Use diagnostics, project analysis, profiling, recovery points, README generation, and launch assets before demo time.

Providers

Use the model that fits the build.

CodeThon keeps builders out of model lock-in. Use hosted frontier APIs, free-tier open models, or local runtimes when privacy and offline work matter.

OpenAI

general coding and tool use

Anthropic

long-context reasoning

NVIDIA

free-tier open models

Groq

fast hosted inference

DeepSeek

reasoning-heavy builds

Together AI

open-model catalog

Ollama

private local runtime

LM Studio

local OpenAI-compatible server

Terminal UX

Live output that explains what the agent is doing.

Generation commands stream formatted markdown. Analysis and profiling commands print each stage before the result. The goal is simple: builders should never wonder whether the CLI froze, hid output, or dumped raw JSON by accident.

Plan the MVP before coding

Generate architecture with context

Stream readable model output

Analyze existing repositories

Profile risky code paths

Create launch and demo assets

CodeThon > /analyze

[run] Analysis: CodeThon-CLI

Scanning file tree in CodeThon-CLI

Reading config and entry files

Detecting stack, entry points, and missing files

Running static project checks

[stream] AI Summary

This project contains a Node.js CLI, provider routing, terminal rendering, guarded tool execution, recovery flows, and a Next.js website.

Safety

Agentic power with visible controls.

CodeThon can write files and run commands, so safety is a core product surface. Builders can use ask mode, dry-run mode, provider tests, doctor checks, and project recovery points while the agent works.

Allowlisted shell tools

Blocked destructive patterns

Filtered child-process environment

Ask and dry-run modes

OS-backed credential storage

Project-local recovery points

Open Source

A terminal agent builders can inspect and extend.

The repository keeps command handlers, provider integrations, terminal UI, runtime policy, recovery systems, tests, and website source open. The npm package ships a compact bundled CLI for global install.

View source npm package Read docs

Release checks

TypeScript build passes

69 Vitest checks pass

npm pack dry-run passes

Installed ct binary smoke-tested

Clean-home setup path verified

Doctor passes on monorepo layout

FAQ

Questions builders ask before using CodeThon CLI.

What is CodeThon CLI?

CodeThon CLI is an open-source AI builder agent that runs in the terminal. It helps builders configure a model, initialize a project, plan architecture, execute implementation tasks, analyze code, profile risks, and recover project context.

Who is it built for?

It is built for hackathon teams, indie builders, startup founders, open-source maintainers, and developers who want an agentic workflow without being locked to one AI provider.

Which AI providers does it support?

CodeThon supports OpenAI, Anthropic, NVIDIA, Groq, DeepSeek, Together AI, Ollama, and LM Studio or other OpenAI-compatible local servers.

How does the terminal workflow start?

Install with npm, run ct, connect a provider, run /init, then use /plan and /execute to move from idea to implementation.

Start Building

Install the CLI, run `ct`, then begin with `/init`.

Use CodeThon as a builder cockpit: plan, execute, debug, profile, recover, and prepare your launch without leaving the terminal.

npm install -g codethon-cli