tethercode.
Get early access

tethercode / models and hosting

Every model behind one address, on machines you run

The work your agents hand to tethercode goes through one router to the models you have connected: the most private that may see it first, then the cheapest, with every call on the record. All of it runs where you put it.

Models and money

The work your agents hand to tethercode goes through one router to the models you have connected: the most private that may see it first, then the cheapest, with every call on the record.

One call, routed
Asked for
a fast model that can read a long file
May see it
three of your five models
Tried first
your local model, the most private
Sent to
your local model, 1,840 tokens in, $0.00
Recorded
who, which agent, which model, how long

Which models may see a call is decided by what is in it.

One address for every model

Agents ask for a kind of model and the router finds a machine that has it. OpenAI, Azure, Google, Mistral, Groq, OpenRouter and DeepInfra, your own vLLM, llama.cpp or LM Studio, and anything else that speaks the same common interface.

Most private first, then cheapest

Among the models allowed and able, the most private is tried first and paid ones are ranked by price. A model that keeps failing is rested, re-checked on its own and brought back without anyone restarting anything.

Every call on the record

Tokens, model, where it ran, for whom, how long, and for paid models the price at that moment. It is a record you can read per day, not a guess at the end of the month.

Long inputs are cut to what matters, not cut off

Text too long for a model is reduced to the parts that bear on the goal. Text that already fits passes through untouched. Answers can be held to an exact shape, and the reply says whether that was enforced or only checked.

Research that cites its sources

Web search and page fetching in every edition, through a search engine you host first. Fetched pages are treated as untrusted text, and addresses inside your network are refused. Premium adds planned, multi-step research that ends in a report with its sources, in Markdown or Word.

Yours to run

One install on your own machine. Your agents connect to it as a set of tools and keep working the way they do now.

What the console shows you
  • Agent board
  • Agent mailbox
  • Approvals
  • Code map
  • Repositories
  • Knowledge
  • Conversations
  • Research and its spend
  • Reasoning review
  • What's missing
  • Artifacts
  • Models and usage
  • Access
  • Secrets

In your browser, served from your own machine.

One command on Windows or Mac

The individual install is a single command on Windows 11 or macOS, running inside Docker Desktop or Podman Desktop. It comes back after a reboot and upgrades with the same command.

One front door for every agent

Agents connect to one address and find several hundred tools there: memory, the code map, tasks, mail, research, models. If the door restarts, their sessions survive it.

You can watch it work

A console shows the agents and their tasks as they happen, the mail between them, what is waiting for your approval, the code map, the knowledge it holds and what the models cost.

Nothing phones home

No usage data is sent to us. What leaves your machine is what you point outward: the models and search you connect, and the licence renewal, which can also be done by carrying a file in.

Releases you can check and undo

Every release is signed, so you can tell who built it and that no file changed. An upgrade backs up the database first, and a rollback puts it back.

It looks after itself

One command reports the health of every service. Services restart themselves, its internal keys are renewed daily, and work that could not be finished waits and retries instead of being dropped.

All of it runs on one machine, for free, on up to three repositories. Get early access

More of tethercode

Try it on your own repository.

Free for 3 repositories, with no time limit and no card. One email when it is ready, and nothing else.

Deploying for a company? Talk to us. Prices are on the editions table.