# Set up a free AI engine

Source: https://career-ops.org/docs/free-ai-engine (canonical HTML, identical content)

> Run career-ops for $0 — no Claude subscription or any paid plan required. Free AI engines for career-ops — OpenCode with a free provider, a local model via Ollama, any OpenAI-compatible endpoint, or the built-in OpenRouter runner — and what to do when you run out of tokens mid-search.



**No, you do not need a Claude subscription — or any paid plan — to run career-ops.** It is free and open source (MIT), and its AI engine can be free too: OpenCode with a free provider, a fully local model via Ollama, any OpenAI-compatible endpoint, or the built-in OpenRouter runner. If you have run out of tokens mid-search, switching to a free engine takes minutes and leaves your CV, pipeline and reports untouched.

career-ops has no paid plans, no tiers, no trials — it is 100% free and MIT-licensed. Any site describing career-ops pricing is describing a different product. The one thing it needs to think is an **AI engine** — an AI coding assistant that reads its prompts and does the work. That engine can be free too.

You bring your own engine and your own key. Nothing runs on our servers — everything happens on your machine. This page shows the free paths, from the easiest to the most hands-on. Pick one and you're ready for the [Quick Start](https://career-ops.org/docs.md).

> **What am I actually setting up?**
Think of career-ops as the recipe and the AI engine as the cook. The recipe is free forever. This page is about getting a cook for free — whether that's a hosted model with a free tier or a model running entirely on your own computer.

## Which path is right for me? [#which-path-is-right-for-me]

| If you want…                      | Use                            | Free?         | Needs a good computer? |
| --------------------------------- | ------------------------------ | ------------- | ---------------------- |
| The easiest free start            | **OpenCode + a free provider** | Yes           | No                     |
| Everything 100% offline & private | **Ollama (local model)**       | Yes           | Yes — 16GB+ VRAM       |
| To reuse a CLI you already have   | **Its own free tier**          | If it has one | No                     |
| The one-command shortcut          | **`npm run or`**               | Yes           | No                     |

Most people should start with the first row. The rest are here when you want them.

## Path 1 — OpenCode + a free provider (recommended) [#path-1--opencode--a-free-provider-recommended]

[OpenCode](https://opencode.ai) is a free, open-source AI coding assistant. Santiago's take: *"opencode is top."* It connects to plenty of providers that offer a **free tier**, so you can run career-ops without spending anything.

**Install OpenCode**
Follow the one-line install on [opencode.ai](https://opencode.ai). It runs in your terminal, like the other AI CLIs.

**Pick a free provider**
Sign up for any provider with a free tier and copy your API key. Popular free options the community uses: **OpenRouter** (models tagged `:free`), **Google AI Studio** (Gemini free tier), and OpenAI-compatible endpoints like **Nvidia's** free tier. The [budget guide](https://github.com/santifer/career-ops/blob/main/docs/RUNNING_ON_A_BUDGET.md) keeps the up-to-date list of models that hold up well.

**Point OpenCode at it**
Set your key and base URL as environment variables, then open OpenCode inside the career-ops folder:

```bash title="Terminal"
export OPENAI_API_BASE="https://openrouter.ai/api/v1"
export OPENAI_API_KEY="your_free_api_key_here"
opencode
```

That's it — OpenCode is now your engine. Head to the [Quick Start](https://career-ops.org/docs.md).

## Path 2 — A CLI you already have [#path-2--a-cli-you-already-have]

Already using a [supported CLI](https://career-ops.org/docs/supported-clis.md) — **Claude Code, OpenCode, Codex, GitHub Copilot CLI, and more**? You're done — career-ops works with all of them, and several ship a free tier. Just open your CLI inside the career-ops folder. No extra setup on this page.

## Path 3 — 100% local with Ollama (fully offline) [#path-3--100-local-with-ollama-fully-offline]

Want zero cloud, zero cost, and total privacy? Run the model on your own machine with [Ollama](https://ollama.com). Nothing leaves your computer.

The trade-off is hardware. career-ops asks the model to score jobs across many dimensions and tailor your CV — small models struggle with that.

> **Pick a model that's big enough**
Skip the tiny 7–8B models — they fail the scoring format and produce generic CVs. Use a **32B model or larger** (e.g. Qwen 2.5 Coder 32B), which needs a GPU with **16–24GB of VRAM** (an RTX 3090/4090, or an Apple Silicon Mac with 32GB+ unified memory). No such machine? Path 1 is the better free route.

```bash title="Terminal"
# Install Ollama from ollama.com, then pull a capable model:
ollama pull qwen2.5-coder:32b
```

Then point your AI CLI (OpenCode works well here) at the local Ollama endpoint, or use the built-in local evaluator described in the [budget guide](https://github.com/santifer/career-ops/blob/main/docs/RUNNING_ON_A_BUDGET.md).

## Path 4 — The one-command shortcut: `npm run or` [#path-4--the-one-command-shortcut-npm-run-or]

career-ops ships a built-in runner that routes to **OpenRouter's free models with automatic fallback** — no CLI configuration to fiddle with. Once you've cloned career-ops and set an OpenRouter key, one command runs the pipeline:

```bash title="Terminal"
export OPENROUTER_API_KEY="your_free_key_here"
npm run or            # runs the full pipeline on free models
```

There are focused variants too — `npm run or:scan`, `or:eval`, `or:apply` — for running a single step.

***

> **For developers — tune models & cost**
Want the maintained model recommendations, per-provider examples, standalone evaluators (`node openai-eval.mjs`, `node ollama-eval.mjs`), and token-saving flags? The core repo's **[Running on a Budget](https://github.com/santifer/career-ops/blob/main/docs/RUNNING_ON_A_BUDGET.md)** guide goes deep. career-ops is fully model-agnostic — point it at any OpenAI-compatible endpoint with zero code changes.

## You're ready [#youre-ready]

Once your engine is set up, continue to the **[Quick Start](https://career-ops.org/docs.md)** — you'll clone career-ops, add your CV, and run your first free scan.