feat(providers): add DeepInfra provider and curated models (#619)

coworker/providers/matrix.py

@@ -16,7 +16,7 @@
 showing a made-up denominator. Values entered 2026-07-28 from vendor docs; verify alongside
 the id refresh.
 
-Resellers: Together + Fireworks + OpenRouter. TODO: add Groq entries here AND its
+Resellers: Together + Fireworks + OpenRouter + DeepInfra. TODO: add Groq entries here AND its
 descriptor in ``registry.py`` once the current provider surface is tested — deliberately
 deferred to bound how much needs verifying at once.
 """
@@ -228,6 +228,32 @@ class ModelEntry:
     "openrouter:stealth/ox-alpha": ModelEntry(
         "Ox Alpha · via OpenRouter", _AGENTIC, 1_048_576
     ),
+    # DeepInfra curated models (OpenAI-compatible inference). Latest generation:
+    # DeepSeek V4, GLM 5.3, Kimi K3. Pinned snapshots (Flash 0731, Pro 0813) and canonical slugs.
+    "deepinfra:deepseek-ai/DeepSeek-V4-Flash": ModelEntry(
+        "DeepSeek V4 Flash · via DeepInfra", _AGENTIC, 128_000
+    ),
+    "deepinfra:deepseek-ai/DeepSeek-V4-Flash-0731": ModelEntry(
+        "DeepSeek V4 Flash 0731 · via DeepInfra", _AGENTIC, 128_000
+    ),
+    "deepinfra:deepseek-ai/DeepSeek-V4-Pro": ModelEntry(
+        "DeepSeek V4 Pro · via DeepInfra", _AGENTIC, 128_000
+    ),
+    "deepinfra:deepseek-ai/DeepSeek-V4-Pro-0813": ModelEntry(
+        "DeepSeek V4 Pro 0813 · via DeepInfra", _AGENTIC, 128_000
+    ),
+    "deepinfra:zai-org/GLM-5.3-Flash": ModelEntry(
+        "GLM-5.3 Flash · via DeepInfra", _AGENTIC_VISION, 1_000_000
+    ),
+    "deepinfra:zai-org/GLM-5.3": ModelEntry(
+        "GLM-5.3 · via DeepInfra", _AGENTIC, 1_000_000
+    ),
+    "deepinfra:moonshotai/Kimi-K3": ModelEntry(
+        "Kimi K3 · via DeepInfra", _AGENTIC_VISION, 1_000_000
+    ),
+    "deepinfra:moonshotai/Kimi-K2.7-Code": ModelEntry(
+        "Kimi K2.7 Code · via DeepInfra", _AGENTIC, 256_000
+    ),
     # -- cloud accounts (models running in the user's own AWS/GCP) ----------------
     # Bedrock ids carry a family segment (claude/ → native Anthropic path, other/ →
     # Converse) plus AWS's own `-v<n>:<m>` version suffix. Some regions require the

coworker/providers/registry.py

@@ -668,6 +668,14 @@ def _responses_compat(
         recommended_model="z-ai/glm-5.2",
         env_key="OPENROUTER_API_KEY",
     ),
+    _compat(
+        "deepinfra",
+        "DeepInfra",
+        base_url="https://api.deepinfra.com/v1/openai",
+        recommended_model="deepseek-ai/DeepSeek-V4-Flash",
+        env_key="DEEPINFRA_API_KEY",
+        endpoint_help="Prefilled with DeepInfra's official OpenAI-compatible endpoint.",
+    ),
     ProviderDescriptor(
         name="ollama",
         title="Ollama (local models)",

tests/test_provider_router.py

@@ -74,6 +74,22 @@ def __init__(self, **kwargs):
     assert captured["api_key"] == "ollama"  # placeholder, Ollama ignores it
 
 
+def test_build_deepinfra_client_uses_default_base_url(monkeypatch):
+    captured: dict = {}
+
+    class FakeOpenAI:
+        def __init__(self, **kwargs):
+            captured.update(kwargs)
+
+    monkeypatch.setattr("openai.OpenAI", FakeOpenAI)
+    client = build_provider_client(
+        "deepinfra", {"api_key": "di-test"}, secrets=None
+    )
+    client._ensure_client()  # type: ignore[attr-defined]
+    assert captured["base_url"] == "https://api.deepinfra.com/v1/openai"
+    assert captured["api_key"] == "di-test"
+
+
 # -- router routing -------------------------------------------------------------
 class _Recorder(ProviderClient):
     def __init__(self, name: str):
@@ -115,6 +131,12 @@ def test_router_routes_and_strips_prefix(monkeypatch):
         "llama3.3"
     ]  # prefix stripped before delegating
 
+    turn = router.complete(
+        model="deepinfra:deepseek-ai/DeepSeek-V4-Flash", messages=[]
+    )
+    assert turn.text == "deepinfra"
+    assert state["latest"]["deepinfra"].models == ["deepseek-ai/DeepSeek-V4-Flash"]
+
     router.complete(model="gpt-5.5", messages=[])  # bare → default openai
     assert state["latest"]["openai"].models == ["gpt-5.5"]
 

tests/test_provider_verify.py

@@ -161,3 +161,13 @@ def test_verify_unexpected_status(monkeypatch):
     res = verify_provider_key("anthropic", api_key="sk-ant-x")
     assert res["ok"] is False
     assert "500" in res["error"]
+
+
+def test_verify_deepinfra_default_endpoint(monkeypatch):
+    cap: dict = {}
+    _patch_get(monkeypatch, status=200, capture=cap)
+
+    assert verify_provider_key("deepinfra", api_key="di-key") == {"ok": True}
+    assert cap["url"] == "https://api.deepinfra.com/v1/openai/models"
+    assert cap["headers"]["Authorization"] == "Bearer di-key"
+

tests/test_providers.py

@@ -498,21 +498,30 @@ def test_matrix_answers_capabilities_for_reseller_ids():
         "fireworks:accounts/fireworks/models/kimi-k2p6",
         "openrouter:z-ai/glm-5.2",
         "openrouter:meta-llama/llama-4-maverick",
+        "deepinfra:deepseek-ai/DeepSeek-V4-Flash",
+        "deepinfra:zai-org/GLM-5.3-Flash",
     ):
         caps = capabilities_for(mid)
         assert caps.tools and caps.parallel_tool_calls and caps.streaming
 
+    assert capabilities_for("deepinfra:zai-org/GLM-5.3-Flash").vision is True
+    assert capabilities_for("deepinfra:deepseek-ai/DeepSeek-V4-Flash").vision is False
+
 
 def test_matrix_labels_and_custom_model_fallback():
     from coworker.providers.matrix import MATRIX, model_labels
 
     labels = model_labels()
     assert labels["together:zai-org/GLM-5.2"] == "GLM-5.2 · via Together"
     assert labels["zai:glm-5.2"] == "GLM-5.2 · Z AI"
+    assert (
+        labels["deepinfra:deepseek-ai/DeepSeek-V4-Flash"]
+        == "DeepSeek V4 Flash · via DeepInfra"
+    )
     # Deliberately small: agent-capable current models only (owner call, 2026-07-04).
-    # 60→65 (2026-08-24): the stealth ox-alpha preview slug tipped it; reclaim slack by
-    # pruning retired entries before raising this again.
-    assert len(MATRIX) < 65
+    # 60→65 (2026-08-24): the stealth ox-alpha preview slug tipped it;
+    # 65→75 (2026-09-04): added DeepInfra curated catalog (DeepSeek V4, GLM 5.3, Kimi K3).
+    assert len(MATRIX) < 75
     assert all(e.caps.tools for e in MATRIX.values())
     # A custom (unlisted) reseller model falls back to the conservative default — usable,
     # but at the user's own risk (no parallel tool calls assumed).
@@ -526,7 +535,7 @@ def test_reseller_descriptors_and_matrix_stay_in_lockstep():
     from coworker.providers.matrix import models_for_provider
     from coworker.providers.registry import get_descriptor
 
-    for name in ("together", "fireworks", "openrouter"):
+    for name in ("together", "fireworks", "openrouter", "deepinfra"):
         d = get_descriptor(name)
         assert d is not None and d.needs_key
         curated = models_for_provider(name)
@@ -536,6 +545,16 @@ def test_reseller_descriptors_and_matrix_stay_in_lockstep():
         assert base.default.startswith("https://")
 
 
+def test_deepinfra_never_leaks_the_openai_key(monkeypatch):
+    import pytest
+    from coworker.providers.registry import build_provider_client
+
+    monkeypatch.setenv("OPENAI_API_KEY", "sk-openai-real")
+    monkeypatch.delenv("DEEPINFRA_API_KEY", raising=False)
+    with pytest.raises(RuntimeError, match="DeepInfra"):
+        build_provider_client("deepinfra", {}, None)
+
+
 def test_foreign_sidecars_stripped_from_outbound_messages():
     """Provider-private sidecars (`_gemini` thought signatures et al) must never reach the
     OpenAI wire — it and its compat servers reject unknown message fields."""

tests/test_token_usage.py

@@ -359,4 +359,12 @@ def test_model_context_windows_covers_verified_entries_only():
     windows = model_context_windows()
     assert windows["anthropic:claude-fable-5"] == 1_000_000
     assert "together:thinkingmachines/Inkling" not in windows  # unverified stays absent
+    assert windows["deepinfra:deepseek-ai/DeepSeek-V4-Flash"] == 128_000
+    assert windows["deepinfra:deepseek-ai/DeepSeek-V4-Flash-0731"] == 128_000
+    assert windows["deepinfra:deepseek-ai/DeepSeek-V4-Pro"] == 128_000
+    assert windows["deepinfra:deepseek-ai/DeepSeek-V4-Pro-0813"] == 128_000
+    assert windows["deepinfra:zai-org/GLM-5.3-Flash"] == 1_000_000
+    assert windows["deepinfra:zai-org/GLM-5.3"] == 1_000_000
+    assert windows["deepinfra:moonshotai/Kimi-K3"] == 1_000_000
+    assert windows["deepinfra:moonshotai/Kimi-K2.7-Code"] == 256_000
     assert all(isinstance(v, int) and v > 0 for v in windows.values())