Questions
I already use Claude Code or Codex. Why would I need this?
Because each of them only knows itself. The moment you run two agents, or come back tomorrow, or want a cheaper model to do the easy work, you become the one carrying context and refereeing. tethercode does that carrying and refereeing, and your agents keep working the way they already do. It also does their searching for them, over an index of your code and documents, so they stop guessing about what they did not read.
Is tethercode an AI model?
No. It has no model of its own for chat or coding. It is the memory, the task ledger, the mailboxes, the code map, the router and the access rules that sit around the agents and models you choose.
What is Kev?
Kev is the small model tethercode uses as a judge. It reads each model call and says how sensitive the content is, and it reads each action an agent is about to take and says whether it could cause harm. It was trained for those decisions and does not chat. On Enterprise it runs on your own GPU; on Free and Premium it is one of the four hosted helper models.
Which agents does it work with?
Claude Code, Codex, Antigravity and Pi, plus models you run locally and any provider you have a key for. They connect to tethercode as a set of tools, so each agent keeps working the way it already does.
Which programming languages does it understand?
Rust, Python, TypeScript, JavaScript, Go, C and C++ are read in depth: what is defined where, what calls what, and what depends on what. About fifty more languages and file types, from Java, C#, Kotlin and Swift to COBOL and Fortran, are read through a general reader. Documents, settings files and database schemas are indexed alongside the code.
What kinds of document can it take in?
PDF, including scanned ones, Word, PowerPoint, Excel, CSV, web pages, and Markdown, JSON, YAML and other plain-text formats. The original file is kept, each revision is a version, and deleting a document removes what was learned from it.
Does my code leave my machine?
tethercode itself runs on your machine and keeps its code map, memory and ledger in your own database. A model call leaves only when the model it is sent to is not local, and the router only sends a call to models you have allowed for content of that sensitivity. On Free and Premium, the four helper models are hosted, so the text they work on goes to whichever provider you pick for them.
Do I need a GPU?
Not for Free or Premium. The helper models that need one are hosted. Enterprise runs everything on the company's own GPU servers.
What does it cost?
Free is $0 for up to 3 repositories, with no time limit. Premium is $19.99 a month and includes 500 million input tokens a month on our hosted helper models. Past that, and on Free, those cost $30 per billion input tokens, paid from a deposit, or you can use another provider. Enterprise is priced per deployment.
What does it run on?
Free and Premium run on one computer: Linux directly, Windows and macOS through Docker Desktop or Podman Desktop. Enterprise runs on a central Linux server, with a small client on each employee's Windows, macOS or Linux machine.
When can I use it?
The product is built and the installers are in final testing. Leave your email and you get one message when it is ready to download.
What do you do with my email address?
We store the address, the edition you picked and which version of this page you were reading. We send one email when tethercode is ready, and the benchmark results if you asked for them. We do not share it or add you to anything else.