Open source · Apache-2.0
All your coding subscriptions.
One terminal app.
Open-source CLI agent. Sign in with the plans you already pay for, keep several logged in at once, and let the harness — goals, loops, schedules, background agents — drive even a cheap model like a senior engineer.
$ bun install -g @hasna/codewith
or npm install -g @hasna/codewith · then run codewith
The harness is the product
A cheap model on its own forgets, drifts, and quits early. Wrapped in Codewith's harness, it sets goals, keeps looping, spawns helpers and works while you're gone — and you only pay small-model prices for all of it.
Autonomous goals
Set a goal and a token budget with /goal. The agent drives itself toward it — pausing, resuming and reporting Goal achieved without babysitting.
Loops that don't quit
/loop every 10 minutes reruns a prompt on this thread forever. Point a cheap model at it and it grinds through the backlog while you sleep.
Schedules
One-time and timed runs with /schedule — /schedule tomorrow at 9am ships the report before standup.
Sub-agents & background agents
Fan work out to durable background agents with /agent — each on its own model. Run the loop on MiniMax while you stay on a flagship.
Many subscription logins
Keep ChatGPT, Claude.ai, Cursor and Grok profiles side by side. Hit a weekly limit? /profile to the next one — same thread, same context.
Switch model mid-session
Change model and provider in one command with /model. Draft cheap, finish sharp — without losing your thread.
Monitors
Watch builds, logs and commands with /monitor and react the moment something changes.
Sandboxed by default
Seatbelt on macOS, Landlock + bubblewrap on Linux, sandboxing on Windows. A cheap model going off-script still can't escape.
Sign in with what you already pay for
No new bills. Bring your coding subscriptions, keep several profiles logged in, and switch between them mid-thread.
● Use your coding subscriptions
OpenAI · ChatGPT plans
Alibaba Qwen
MiniMax
Kimi
Z.ai
Xiaomi MiMo
Bring your own key
Anthropic
Google Gemini
xAI
DeepSeek
OpenRouter
Cerebras
NVIDIA
Amazon Bedrock
Run local
Ollama
LM Studio
…or point a custom provider at any OpenAI-compatible endpoint.
Yours to read, run, and fork
Apache-2.0 licensed and developed in the open. The Bun/npm package installs the right macOS, Linux, or Windows binary automatically. Source archives and manual release binaries stay available on GitHub Releases for people who want them.
View on GitHub