Linux

This guide gets the repo-local vex binary running on Linux.

Before you start:

use a checkout where you can write .vex/ and AGENTS.md; vex init creates .vex/config.toml, .vex/validate.toml, and AGENTS.md
vex needs an external model endpoint; it does not start a model server for you
Git and a working C toolchain must already be available through your distro packages
local and private-network endpoints do not need VEX_MODEL_TOKEN; remote public endpoints do

1. Verify Git and the C toolchain, then install Rust

git --version
cc --version
curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh -s -- -y
source "$HOME/.cargo/env"
rustc --version
cargo --version

2. Clone the repository

git clone https://github.com/aistar-au/vexcoder.git
cd vexcoder

3. Build the release binary

cargo build --release

4. Scaffold the repo-local workspace files

./target/release/vex init

5. Point `vex` at a local server

The fastest local setup is a server that listens on http://127.0.0.1:8080/v1.

cat > .vex/config.toml <<'EOF'
model_url = "http://127.0.0.1:8080/v1"
model_name = "local/default"
model_profile = "models/local-balanced.toml"
EOF

If you prefer scheme-and-host discovery instead of a fixed model_url, see Configuration and set api_client.base_url instead.

6. Start your local server in another shell session

Make sure it is already listening on 127.0.0.1:8080 before the next step. If your server uses a different host or port, update model_url to match.

7. Verify the endpoint and permissions

vex needs normal user-level write access in this checkout and network access to model_url. No extra privileges are required after the toolchain install completes.

./target/release/vex doctor

Expect VEX_MODEL_URL set to pass. Model endpoint reachable should pass once your server is listening; if it warns, start your local server or update model_url, then rerun vex doctor.

8. Remote endpoints require HTTPS and a token

If you switch away from the local-server path, set model_url to the exact https:// endpoint your server exposes, set model_name to a model that endpoint accepts, and then export the token before rerunning vex doctor.

export VEX_MODEL_TOKEN="your-token"

9. Send one request and open the interactive app

./target/release/vex --project-map-only "Reply with the single word ok."
./target/release/vex

VexCoder