Add Image Support - thanks, @montasaurus

* add image support
* bump python and llm versions
* reset user directory on each test
* add recordings to readme
* Ensure compatible OpenAI library

---------

Co-authored-by: Simon Willison <swillison@gmail.com>

Author: montasaurus

.github/workflows/publish.yml

@@ -12,7 +12,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
+        python-version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
     steps:
     - uses: actions/checkout@v4
     - name: Set up Python ${{ matrix.python-version }}
@@ -38,7 +38,7 @@ jobs:
     - name: Set up Python
       uses: actions/setup-python@v5
       with:
-        python-version: "3.12"
+        python-version: "3.13"
         cache: pip
         cache-dependency-path: pyproject.toml
     - name: Install dependencies

.github/workflows/test.yml

@@ -10,7 +10,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python-version: ["3.8", "3.9", "3.10", "3.11", "3.12"]
+        python-version: ["3.9", "3.10", "3.11", "3.12", "3.13"]
     steps:
     - uses: actions/checkout@v4
     - name: Set up Python ${{ matrix.python-version }}
@@ -25,4 +25,3 @@ jobs:
     - name: Run tests
       run: |
         pytest
-

README.md

@@ -54,6 +54,13 @@ Now you can prompt Claude using:
 ```bash
 cat llm_openrouter.py | llm -m claude -s 'write some pytest tests for this'
 ```
+
+Images are supported too, for some models:
+```bash
+llm -m openrouter/anthropic/claude-3.5-sonnet 'describe this image' -a https://static.simonwillison.net/static/2024/pelicans.jpg
+llm -m openrouter/anthropic/claude-3-haiku 'extract text' -a page.png
+```
+
 ## Development
 
 To set up this plugin locally, first checkout the code. Then create a new virtual environment:
@@ -64,9 +71,13 @@ source venv/bin/activate
 ```
 Now install the dependencies and test dependencies:
 ```bash
-pip install -e '.[test]'
+llm install -e '.[test]'
 ```
 To run the tests:
 ```bash
 pytest
 ```
+To add new recordings and snapshots, run:
+```bash
+pytest --record-mode=once --inline-snapshot=create
+```

llm_openrouter.py

@@ -29,10 +29,12 @@ def register_models(register):
     if not key:
         return
     for model_definition in get_openrouter_models():
+        supports_images = get_supports_images(model_definition)
         register(
             OpenRouterChat(
                 model_id="openrouter/{}".format(model_definition["id"]),
                 model_name=model_definition["id"],
+                vision=supports_images,
                 api_base="https://openrouter.ai/api/v1",
                 headers={"HTTP-Referer": "https://llm.datasette.io/", "X-Title": "LLM"},
             )
@@ -78,3 +80,14 @@ def fetch_cached_json(url, path, cache_timeout):
             raise DownloadError(
                 f"Failed to download data and no cache is available at {path}"
             )
+
+
+def get_supports_images(model_definition):
+    try:
+        # e.g. `text->text` or `text+image->text`
+        modality = model_definition["architecture"]["modality"]
+
+        input_modalities = modality.split("->")[0].split("+")
+        return "image" in input_modalities
+    except Exception:
+        return False

pyproject.toml

@@ -9,8 +9,9 @@ classifiers = [
     "License :: OSI Approved :: Apache Software License"
 ]
 dependencies = [
-    "llm>=0.8",
-    "httpx"
+    "llm>=0.17",
+    "httpx",
+    "openai>=1.57.0",
 ]
 
 [project.urls]
@@ -23,4 +24,4 @@ CI = "https://github.com/simonw/llm-openrouter/actions"
 openrouter = "llm_openrouter"
 
 [project.optional-dependencies]
-test = ["pytest", "pytest-httpx"]
+test = ["pytest", "pytest-recording", "inline-snapshot"]

tests/cassettes/test_llm_openrouter/test_image_prompt.yaml

@@ -1,30 +1,5 @@
-import json
-import llm
 import pytest
 
-DUMMY_MODELS = {
-    "data": [
-        {
-            "id": "openai/gpt-3.5-turbo",
-            "pricing": {"prompt": "0.0000015", "completion": "0.000002"},
-            "context_length": 4095,
-            "per_request_limits": {
-                "prompt_tokens": "2871318",
-                "completion_tokens": "2153488",
-            },
-        },
-        {
-            "id": "anthropic/claude-2",
-            "pricing": {"prompt": "0.00001102", "completion": "0.00003268"},
-            "context_length": 100000,
-            "per_request_limits": {
-                "prompt_tokens": "390832",
-                "completion_tokens": "131792",
-            },
-        },
-    ]
-}
-
 
 @pytest.fixture
 def user_path(tmpdir):
@@ -36,7 +11,11 @@ def user_path(tmpdir):
 @pytest.fixture(autouse=True)
 def env_setup(monkeypatch, user_path):
     monkeypatch.setenv("LLM_USER_PATH", str(user_path))
-    # Write out the models.json file
-    (llm.user_dir() / "openrouter_models.json").write_text(
-        json.dumps(DUMMY_MODELS), "utf-8"
+    monkeypatch.setenv(
+        "LLM_OPENROUTER_KEY",
+        "sk-...",
+    )
+    monkeypatch.setenv(
+        "OPENROUTER_KEY",
+        "sk-...",
     )

tests/cassettes/test_llm_openrouter/test_llm_models.yaml

@@ -1,22 +1,71 @@
+import llm
+import pytest
 from click.testing import CliRunner
+from inline_snapshot import snapshot
 from llm.cli import cli
-import json
-import pytest
+
+TINY_PNG = (
+    b"\x89PNG\r\n\x1a\n\x00\x00\x00\rIHDR\x00\x00\x00\xa6\x00\x00\x01\x1a"
+    b"\x02\x03\x00\x00\x00\xe6\x99\xc4^\x00\x00\x00\tPLTE\xff\xff\xff"
+    b"\x00\xff\x00\xfe\x01\x00\x12t\x01J\x00\x00\x00GIDATx\xda\xed\xd81\x11"
+    b"\x000\x08\xc0\xc0.]\xea\xaf&Q\x89\x04V\xe0>\xf3+\xc8\x91Z\xf4\xa2\x08EQ\x14E"
+    b"Q\x14EQ\x14EQ\xd4B\x91$I3\xbb\xbf\x08EQ\x14EQ\x14EQ\x14E\xd1\xa5"
+    b"\xd4\x17\x91\xc6\x95\x05\x15\x0f\x9f\xc5\t\x9f\xa4\x00\x00\x00\x00IEND\xaeB`"
+    b"\x82"
+)
+
+
+@pytest.mark.vcr
+def test_prompt():
+    model = llm.get_model("openrouter/openai/gpt-4o")
+    response = model.prompt("Two names for a pet pelican, be brief")
+    assert str(response) == snapshot("Gully or Skipper")
+    response_dict = dict(response.response_json)
+    response_dict.pop("id")  # differs between requests
+    assert response_dict == snapshot(
+        {
+            "content": "Gully or Skipper",
+            "role": "assistant",
+            "finish_reason": "stop",
+            "usage": {"completion_tokens": 5, "prompt_tokens": 17, "total_tokens": 22},
+            "object": "chat.completion.chunk",
+            "model": "openai/gpt-4o",
+            "created": 1731200404,
+        }
+    )
 
 
-@pytest.mark.parametrize("set_key", (False, True))
-def test_llm_models(set_key, user_path):
+@pytest.mark.vcr
+def test_llm_models():
     runner = CliRunner()
-    if set_key:
-        (user_path / "keys.json").write_text(json.dumps({"openrouter": "x"}), "utf-8")
     result = runner.invoke(cli, ["models", "list"])
     assert result.exit_code == 0, result.output
     fragments = (
         "OpenRouter: openrouter/openai/gpt-3.5-turbo",
         "OpenRouter: openrouter/anthropic/claude-2",
     )
     for fragment in fragments:
-        if set_key:
-            assert fragment in result.output
-        else:
-            assert fragment not in result.output
+        assert fragment in result.output
+
+
+@pytest.mark.vcr
+def test_image_prompt():
+    model = llm.get_model("openrouter/anthropic/claude-3.5-sonnet")
+    response = model.prompt(
+        "Describe image in three words",
+        attachments=[llm.Attachment(content=TINY_PNG)],
+    )
+    assert str(response) == snapshot("Red and green")
+    response_dict = response.response_json
+    response_dict.pop("id")  # differs between requests
+    assert response_dict == snapshot(
+        {
+            "content": "Red and green",
+            "role": "assistant",
+            "finish_reason": "end_turn",
+            "usage": {"completion_tokens": 7, "prompt_tokens": 82, "total_tokens": 89},
+            "object": "chat.completion.chunk",
+            "model": "anthropic/claude-3.5-sonnet",
+            "created": 1731200406,
+        }
+    )