# Quickstart

The shortest path from nothing to a working agent: one install command, one subscription sign-in, one model. No API key, no config file editing.

If you already have a provider account, [Getting Started](/getting-started) covers the same ground with more detail and more options.

## 1. Install

```bash
curl -fsSL https://agenvoy.com/scripts/install.sh | bash
```

The script detects macOS or Linux, installs the `agen` binary to `/usr/local/bin`, and pulls in what it needs (`git`, `make`, a C compiler, `python3`, `node`, `pdftotext`, and `bwrap` on Linux). It asks for `sudo` once, to write into `/usr/local/bin`.

Check it landed:

```bash
command -v agen
```

## 2. Subscribe to GitHub Copilot

GitHub Copilot ($10/mo) is the lowest-cost starting point Agenvoy recommends: the price is fixed monthly, so an agent that takes a few extra turns doesn't move the bill.

Agenvoy no longer recommends a free plan. Since 2026-11-09 (5:40 PM PT), Ollama Cloud's Free plan is pay-as-you-go, and its hosted models (`gemma4:31b`, `gpt-oss:120b`, `gpt-oss:20b`, `nemotron-3-super`, `nemotron-3-ultra`) need purchased credits. Running models locally with Ollama stays free, but small local models call tools unreliably.

1. Open **[github.com/features/copilot/plans](https://github.com/features/copilot/plans)**
2. Sign in and subscribe to the $10/mo plan

## 3. Add the model

Start the TUI:

```bash
agen
```

Then, inside it:

| Step | What you do |
|---|---|
| 1 | Type `/model` and press **Enter** |
| 2 | Choose **add** — "add model from provider" |
| 3 | The picker opens on the **OAuth** tab. Choose **GitHub Copilot** |
| 4 | The popup shows a verification URL and a user code; press **Enter** to open that page in the browser, enter the code and approve within 15 minutes. The token goes straight to the OS keychain — never into a config file, never into the conversation |
| 5 | Pick **`gpt-5.6-luna`** from the models your Copilot plan offers |

That is the whole setup. The model is registered in `~/.config/agenvoy/config.json`; the credential is not.

> Pick `gpt-5.6-luna` because its quota drains slowly enough for daily use and it is capable enough for most work. **Shift+U** shows how much Copilot quota is left.

## 4. Optional: add the decision model (TypeSafe Jev)

The decision model is a second, much smaller model that never writes replies. It answers multiple-choice questions for the agent in a fraction of a second: which skills fit this request, which search results and page blocks are noise, which memory entries matter now. Agenvoy runs without it; with it, each turn carries less irrelevant context. Jev is TypeSafe's model for this job.

1. Open **[console.typesafe.ai/keys](https://console.typesafe.ai/keys)**, sign in and create an API key. Jev is in early access and is billed per input token only: $42 per billion input tokens, output free ([docs.typesafe.ai/models](https://docs.typesafe.ai/models), checked 2026-11-11)
2. In the TUI, type `/model` → **add** → switch to the **API Key** tab → choose **TypeSafe** (pressing `o` on that row opens the same key page) → paste the key. The key goes to the OS keychain as `TYPESAFE_API_KEY`, and the notice reads `TypeSafe API key saved; pick its model in /model decision`
3. Type `/model` → switch to the **decision** tab → choose **`typesafe@jev-latest`**, which follows the newest Jev release

### What it has done since v1.3.0

The decision model arrived in v1.3.1, the first release after v1.3.0, replacing the single-purpose Jev dispatcher beta, and has taken on more checks in every release since ([Decision Model](/providers-typesafe) lists them all). Measured on the author's machine from 2026-11-05 to 2026-11-11, from the daemon's own debug log rather than a benchmark:

| Check | Calls | Result |
|---|---|---|
| Skill list | 576 | 11,353 of 11,846 skill entries (95.8%) left out of the turn context as unrelated, so a turn lists about one skill instead of about 21 |
| Session summary | 32 | 78 memory entries narrowed to the 27 that bore on the request (v1.3.7) |
| Summary merge | 159 | 74 of 496 exchanges dropped as noise and 20 outdated decisions removed before merging |
| Tool gap | 565 | 8 requests flagged as needing a tool that did not exist yet, pointing the agent straight to building one |
| Fetched pages | 60 | 111 of 3,677 page blocks removed as navigation or boilerplate before caching |
| Search results | 60 | 9 of 648 results dropped as off-topic |

Only 2 of those checks failed, and a failed check leaves the content as it was, so the worst case is the behaviour you would get without a decision model.

## 5. Ask it something

Type a question into the input box and press **Enter**:

```
What's the weather in Taipei right now?
```

No weather tool ships with Agenvoy. Watch what happens instead: the agent searches its own registry, finds nothing, fetches the tool-building contract, picks a public API, writes a Python tool, sandbox-tests it, then calls it and answers. The tool is saved — the second time you ask, it just runs.

Try a few more:

```
Summarize the README in this folder
Report the TSMC stock price every morning at 8am
Which of my documents mentions the prompt guide?
```

Next: [After the First Run](/quickstart-next) covers the six keys to know and where to read next.