Skip to content

tui-endpoint-quickstart.md

Latest commit

Β 

History

History
269 lines (197 loc) Β· 9.09 KB

tui-endpoint-quickstart.md

File metadata and controls

269 lines (197 loc) Β· 9.09 KB
Raw

FrontierAgent TUI quickstart for macOS and Linux

δΈ­ζ–‡ Β· Documentation index

This tutorial is for developers who want to run the FrontierAgent TUI on macOS or Linux and connect it to an existing OpenAI-compatible LLM endpoint configured in .env. It does not deploy a model, require an NVIDIA GPU, or require Docker.

What you will run

Your macOS or Linux terminal
β”œβ”€β”€ FrontierAgent full-screen TUI
β”œβ”€β”€ target code repository (selected by --cwd)
└── HTTPS β†’ an existing OpenAI-compatible LLM endpoint

FrontierAgent and the model service are independent. Only the agent and its tools run on your machine; the endpoint configured in .env performs model inference.

1. Prerequisites

You need only:

  • a macOS or Linux terminal;
  • Git;
  • network access to GitHub, Python package sources, and the LLM endpoint;
  • the endpoint's API key, base URL, and model ID.

The project uses uv to manage Python 3.12 and its dependencies. The launcher installs uv from Astral's official site if it is missing. If your organization disallows automatic downloads, ask an administrator to install uv first.

2. Clone the project and configure .env

git clone https://github.com/ApodexAI/FrontierAgent.git
cd FrontierAgent
cp .env.example .env
chmod 600 .env

Open .env in an editor and set at least these three values:

OPENAI_API_KEY=your-api-key
OPENAI_BASE_URL=https://your-endpoint.example/v1
OPENAI_MODEL=your-model-id

Configuration rules:

  • the endpoint must support the OpenAI Chat Completions API;
  • OPENAI_BASE_URL normally ends in /v1; do not include /v1/chat/completions;
  • OPENAI_MODEL must exactly match a model ID accepted by the endpoint;
  • .env is ignored by Git. Never commit, screenshot, or share it;
  • environment variables take precedence over .env, so existing shell values with the same names override the file.

Search and document settings such as SERPER_API_KEY and JINA_API_KEY are optional and are not needed for the first TUI run.

3. Start the TUI

Run these commands from the cloned FrontierAgent directory. Replace /absolute/path/to/your-project with the absolute path of the repository the agent should work on.

macOS

./scripts/run-macos.sh \
  --mode react \
  --cwd /absolute/path/to/your-project

Linux

./scripts/run-linux.sh \
  --mode react \
  --cwd /absolute/path/to/your-project

The launcher installs Python 3.12 and the required dependencies, validates .env, and opens the full-screen TUI. It does not install or start a local LLM. The first run downloads dependencies; later runs reuse the cache.

Default runtime selection is platform-specific:

  • Linux uses the workspace-local native runtime by default;
  • macOS selects automatically: it uses Docker when the daemon is available and falls back to native mode otherwise;
  • pass --native to explicitly require native mode;
  • on Linux, pass --bwrap when bubblewrap is installed and supported for a lightweight filesystem boundary;
  • pass --docker when a reachable Docker daemon is available for container isolation.

Linux runtime selection can be summarized as follows:

Invocation Behavior
No runtime option Defaults to native; does not probe or switch to Docker/bubblewrap
--native Explicitly requires native mode
--bwrap Explicitly requires bubblewrap; startup fails if unavailable
--docker Explicitly requires Docker; startup fails if unavailable

Explicit isolation choices are fail-closed. If bubblewrap or Docker is unavailable, FrontierAgent exits with an error instead of silently falling back to native mode. This prevents a developer from mistakenly believing the task is still isolated.

To install bubblewrap on Debian/Ubuntu and start FrontierAgent:

sudo apt-get update
sudo apt-get install -y bubblewrap
./scripts/run-linux.sh --bwrap --mode react --cwd /absolute/path/to/your-project

Installing the binary alone may not be sufficient: the host or cloud platform must also permit Linux user namespaces. The launcher checks actual usability.

For example, to explicitly use Docker on either platform:

# Docker must be installed and running
./scripts/run-macos.sh --docker --mode react --cwd /absolute/path/to/your-project
./scripts/run-linux.sh --docker --mode react --cwd /absolute/path/to/your-project

Docker is only an isolation boundary for agent commands. FrontierAgent still calls the remote LLM endpoint from .env; it does not deploy a model locally.

Add --setup-only to either command if you want to install dependencies and validate configuration without opening the TUI.

4. Run your first task

When the TUI opens, enter this read-only task in the prompt and press Enter:

Read this repository's README and project configuration. Summarize its purpose, entry points, and local test commands. Analyze only; do not modify files.

Start with react: one stateful agent performs code reading, command execution, and file work in sequence. For work that genuinely splits into independent subtasks, exit and start Agent Team instead:

# macOS
./scripts/run-macos.sh --mode agent_team --cwd /absolute/path/to/your-project

# Linux
./scripts/run-linux.sh --mode agent_team --cwd /absolute/path/to/your-project

agent_team uses a coordinator and sub-agents and will normally consume more endpoint tokens.

5. Essential TUI controls

Control Purpose
/help or F1 Show complete help
/config Show redacted provider, model, and endpoint diagnostics
/mode react Switch to the single-agent workflow
/mode agent_team Switch to Agent Team
/plan Investigate and propose a plan; edits stay locked until approval
/attach <path> Add a read-only file or directory input
/log Show the current run's trace path
/revert Undo changes recorded by file-editing tools in this session
/resume Pick and resume a saved session inside the TUI
Ctrl-C Interrupt the current task
/exit Exit the TUI

Writes show a diff and request approval by default. Do not add --yes on your first run. Native mode is not an operating-system sandbox: approved commands run with the permissions of your current macOS or Linux user. Select --docker explicitly when you want a container isolation boundary.

6. Files and sessions

For a native run started with --cwd /absolute/path/to/your-project, state is stored under:

/absolute/path/to/your-project/.apodex/
β”œβ”€β”€ runs/<session-id>/       # trace, engine log, checkpoint, and outputs
└── runtime/native/          # workspace-local caches, temporary files, and dependencies

List resumable sessions:

uv run frontier-agent --cwd /absolute/path/to/your-project --resume

Resume a specific session ID:

uv run frontier-agent \
  --cwd /absolute/path/to/your-project \
  --resume SESSION_ID

7. Troubleshooting

The full-screen TUI does not open

The TUI requires both stdin and stdout to be attached to a normal TTY. Do not pipe or redirect the launcher, use TERM=dumb, or add --print, --no-tui, --no-color, or --theme mono.

Required .env values are missing

Make sure OPENAI_API_KEY, OPENAI_BASE_URL, and OPENAI_MODEL are non-empty in FrontierAgent/.env, then rerun the launcher from the FrontierAgent root.

401 or 403

The API key is invalid, expired, or lacks access to the selected model. Correct .env and restart FrontierAgent.

404 or model not found

Check that the base URL normally includes /v1 but not /chat/completions, and that the model ID exactly matches one exposed by the endpoint.

Timeout or connection failure

Check endpoint reachability from this machine along with VPN, proxy, DNS, TLS certificate, and firewall policies. FrontierAgent's startup preflight validates local configuration only; it does not make a network request to test the key.

uv is still not found after installation

Open a new terminal, or follow the official uv instructions to add its binary directory to PATH. The launchers also check $HOME/.local/bin/uv and $HOME/.cargo/bin/uv directly.

8. Shortest command for daily use

After the first setup, continue to launch from the FrontierAgent repository:

# macOS
./scripts/run-macos.sh --cwd /absolute/path/to/your-project

# Linux
./scripts/run-linux.sh --cwd /absolute/path/to/your-project

If you would rather type frontier-agent inside the project itself, install the CLI once as a tool and keep the endpoint in the user env file. That setup, its precedence rules, and the extra Docker step on macOS are in Install once and launch from any project.

Next, read the TUI user guide for the three sidebar tabs, Space previews, approvals, and Agent Team asynchronous intervention.

For the lower-level CLI/TUI, approval, attachment, theme, and session reference, see apodex/README.md. For platform details, see the macOS guide and Linux guide.