> ## Documentation Index
> Fetch the complete documentation index at: https://docs.jelou.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# AI models

> Discover and validate the AI models available to the AI Agent and AI_TASK nodes: provider and capability filters, validation, and catalog cache.

`jelou models` queries the catalog of AI models available to your organization —
the ones you can use in the **AI Agent** and **AI\_TASK** nodes. It serves two
purposes: discovering which models you have, and validating that the name you are
about to write into a workflow actually exists.

<Tip>
  If your agent already needs other workspace context (identity, project drift,
  secrets, databases), `jelou context --agent` brings all of that back in a
  single call, including this same model allowlist in `data.models` (with
  `source`, `fetchedAt`, `count`, `allowed[]`, `deprecated[]`, and
  `byProvider{}`). It's a handy shortcut when you're going to request that
  broader context anyway.
</Tip>

## List

```bash theme={null}
jelou models list                                  # all
jelou models list --provider anthropic             # from one provider
jelou models list --capability vision --json       # ones that support images
```

| Flag                                              | Description                                                                                 |
| ------------------------------------------------- | ------------------------------------------------------------------------------------------- |
| `--provider <name>`                               | Filter by provider                                                                          |
| `--capability <cap>`                              | `vision`, `json`, `pdf`, `reasoning`, `streaming`, `audio`, `system`, `tools`, `structured` |
| `--type <type>`                                   | `platform` or `custom` (yours)                                                              |
| `--include-deprecated`                            | Include models with an end-of-life date                                                     |
| `--no-cache`                                      | Fresh fetch, skip the local cache                                                           |
| `--sort <catalog\|name\|input-cost\|output-cost>` | Sort order for the results (defaults to `catalog`)                                          |
| `--max-input-cost <number>`                       | Only models at or below this input cost per 1M tokens                                       |
| `--max-output-cost <number>`                      | Only models at or below this output cost per 1M tokens                                      |

Deprecated models **do not appear** unless you pass `--include-deprecated`.

## Show details

Accepts the short name, the qualified id, or the numeric id:

```bash theme={null}
jelou models show gpt-4.1
jelou models show openai/gpt-4.1
```

## Validate

Confirms a model exists before you write it into a workflow. This is the command
built for CI and for agents:

```bash theme={null}
jelou models validate openai/gpt-4.1        # canonical form
jelou models validate gpt-4.1               # bare form, auto-expanded
jelou models validate gpt-4o --provider openai
```

| Exit code | Meaning                                                    |
| --------- | ---------------------------------------------------------- |
| `0`       | Valid and not deprecated                                   |
| `2`       | Deprecated, or the bare name is ambiguous across providers |
| `3`       | Not in the catalog                                         |

| Flag                 | Description                                                |
| -------------------- | ---------------------------------------------------------- |
| `--provider <name>`  | Require the model to belong to that provider               |
| `--allow-deprecated` | Treat deprecated models as valid (exit `0` with a warning) |

<Tip>
  `--provider` is useful for validating a `fallbackModel`'s model+provider pair
  before publishing a workflow.
</Tip>

## Catalog cache

The catalog is cached on disk so the API isn't queried on every call. If you just
enabled a model and it does not show up:

```bash theme={null}
jelou models refresh                        # fresh fetch, rewrites the cache
jelou models refresh --clear                # wipe the active profile's cache
jelou models refresh --clear --all-profiles # wipe every profile's cache
```

You can also pass `--no-cache` to `list`, `show` and `validate` for a one-off
query that leaves the cache alone.