Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -83,6 +83,12 @@ Nothing. We built T3 Code because we wanted the best possible development experi

We wanted something performant, remote-ready, and truly open. If we ever go the wrong direction, we want you to have everything you need to fork and build the tool that you want.

## Prime Agent

This fork documents how to use Prime Agent as a local ACP provider in T3 Code.
Follow the [Prime Agent setup guide](./docs/user/providers-acp.md#prime-agent)
to add the provider, configure a model, and test tool execution with a read-only file check.

## Documentation

- [Install and first run](./docs/user/install.md)
Expand Down
40 changes: 40 additions & 0 deletions docs/user/providers-acp.md
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,46 @@ credentials and skills remain managed by the installed agent or its wrapper. Mod
options come from ACP, including model IDs that contain JSON. T3 Code launches the executable directly
without expanding shell expressions.

### Prime Agent

Install [Prime Agent](https://github.com/PrimeIntellect-ai/prime-agent/releases) on the machine
hosting the T3 server. Add a local ACP command with the Prime Agent executable and two argument
rows, `--mode` and `acp`. Configure model access in Prime Agent before opening a thread.

For a local Ollama trial, configure Prime Agent's custom provider with the Ollama OpenAI-compatible
endpoint and the exact installed model tag. When both processes run on the same host, use
`http://127.0.0.1:11434/v1`. To pin the launch settings, add separate argument rows for
`--provider`, the configured provider name, `--model`, the tag, `--thinking`, and the desired
level. Use a dedicated profile via the `PRIME_AGENT_CODING_AGENT_DIR` environment override if you
want separate credentials and session state.

On Windows, Prime Agent needs its Python kernel to execute tools. Make `uv` available to the
agent for automatic setup, or set `PRIME_AGENT_KERNEL_PYTHON` to an interpreter with the runtime
bundled with your Prime Agent release and its
[default kernel packages](https://github.com/PrimeIntellect-ai/prime-agent/blob/v0.10.3/crates/pa-core/src/kernel/bootstrap/mod.rs).
For a separate daemon on Windows, use a full named pipe path for `PRIME_AGENT_DAEMON_SOCKET`,
such as `\\.\pipe\prime-agent-t3-lab`.

Prime Agent 0.10.3 advertised a model selector and no reasoning selector for the tested custom
Ollama model. With that configuration, select reasoning using the launch arguments. Test the
provider in a disposable project before giving it access to a working checkout.

#### Try a read-only file check

1. Create a disposable folder with a file named `sample.txt`. Put one line in it, such as
`Fixture is ready.`
2. Add that folder as a T3 project. In a new chat, select your Prime Agent provider and the
configured Ollama model in the model picker.
3. Send this prompt:

> Use one Python tool call to read sample.txt in this folder. Return its exact line.
> Leave the file as it is and use only local tools.

The chat should display a tool action and reply with `Fixture is ready.` Check that the file
still contains its original line. This exercises model selection and tool execution in one
turn. If the chat reaches the tool step and reports a kernel startup error, check the Python
runtime and default packages before retrying.

## Where agents run

Registry agents always run on the machine that hosts your T3 Code server. That stays true when you
Expand Down
Loading