Agents share a sandbox. Open any screenshot for a closer look.
The development lifecycle
Know what you're shipping.
Keep the plan beside the code. You can check the design as your agents work.
01 / Planning
Give your agent a plan.
Keep your team's backlog on one board. You write the queue and choose what comes next. Your agent works through it, one task at a time. Open a sandbox from a task to keep the code tied to the request.
Your agent finishes. You review the work before checking it off.
See what needs review.
02 / Design
Keep the thinking close.
Pick up the reasoning behind the last change. Your project keeps its notes beside its tasks. Write down the decision and sketch how the system fits together.
Check the source without leaving the sandbox.
The decision and the drawing, in one note.
03–04 / Implementation and review
Review what your agents write.
Read the work at your own pace. Agent GUI mode turns the transcript into a timeline. Highlight a passage and leave a comment. Send your comments back to the running agent.
Check the diff in Changes. Ask another agent tab to review the code before you open a pull request.
Codex fixes the parser and checks the result.
A computer for the task
Run the tools you need.
Most agent sandboxes put you in a locked-down container. They leave out or block basic tools. With term.so, you get the whole VM.
Run your projects the way you do on your machine. You get a shell and sudo to install any software. Run Docker and manage services with systemd. Build code from several repos and run your databases in the same sandbox.
Each sandbox runs in its own dedicated microVM. When you park it, your work stays put until you resume.
Model providers
Pick your agents.
Bring your own subscriptions and mix providers in one sandbox. Choose Claude, Codex, Kimi, Antigravity, Pi or OpenCode.
Named tabs
Keep your model costs down.
A stronger reasoning model, like Fable or Astra, plans and reviews the work. Smaller, cheaper models write the code. You save on implementation tokens because the expensive model reads the plan and diff.
Give each tab a job, like “Review” or “Implementation”. One tab steers another with zippy tab send and waits with zippy tab wait. You don't need copy-paste or glue scripts.
Terminal
Make it feel like yours.
Import your iTerm theme to color the terminal and the rest of the app. Pick your font and keybindings.
Your sandbox keeps running when you close the browser. Open it on another device and carry on.
Choose an agent. Give the tab a job.Choose an agent. Give the tab a job.Astra reviews the work and sends feedback to Implementation.Kimi explains how the retry code works.
Workspaces
Bring your team together.
Create a workspace for your team's tasks and notes. Invite the people you work with. Switch workspaces from the sidebar.
Mobile
Take your work with you.
Check the task board on your phone. Open the same sandbox to read the agent's progress and steer what happens next.
Models are rocket engines. Strap one to a horse cart and the process can't keep up. term.so is the rocket shuttle: you can work through the task and question the design as agents write the code.
We built term to make thinking through the work feel natural. That effort builds your judgement, so you can ship systems you understand.
We're a team of tech leads from Google, Facebook, Stripe, Airbnb and Uber.
Pricing
One plan for your team.
Your seat is $24 a month. Sandboxes bill by the vCPU-hour while they run. Parked ones are free. Your model subscriptions stay separate.