Add Google Gemini platform via OpenAI-compat endpoint (provider 7)

The 10th registry row: id "google", GEMINI_API_KEY > auth.json "google" scope, https://generativelanguage.googleapis.com/v1beta/openai (Gemini's OpenAI-compatibility shim) with KIGI_GOOGLE_BASE_URL override, Bearer, OpenAI listing + ChatCompletions + Passthrough, enrichment-backed metadata (models_dev_id google), tool-calling listing restriction. Review caught a ship-blocking defect: Gemini's compat /models returns -PREFIXED ids (confirmed via Google's own cookbook), but the models.dev snapshot keys and the chat endpoint use the BARE id. Without normalization the enrichment lookup misses and restrict_to_enriched silently empties the Gemini catalog (login works, zero models selectable). A doc WebFetch had hidden this — the compat docs show bare INPUT ids to retrieve/chat but never print the list OUTPUT. Fix: new spec field strip_listing_id_prefix, applied in the fetch before filter/enrich/keying. Google sets Some("models/") (defensive: a no-op if an id is already bare, so correct regardless of the live shape); all other rows None. The e2e now feeds the REAL prefixed listing ids and asserts they survive as the bare managed key google/gemini-2.5-pro with the bare id on the wire (chat rejects the prefix); a registry test pins the config. Note: models.dev models Gemini reasoning as budget_tokens (not effort-type), so no auto effort-menu — models still fully work; reasoning is dynamic. Gemini compat applies default safety filters (no BLOCK_NONE).
2026-07-21 11:47:47 -04:00
parent e373ff8b29
commit 7a2cd8a726
4 changed files with 208 additions and 9 deletions
@@ -144,6 +144,14 @@ struct PlatformSpec {
    /// (tts/embeddings/image). Availability still requires the LIVE listing;
    /// this only drops listing noise, never adds models.
    restrict_to_enriched: bool,
+    /// A prefix to strip from each live-listing model id before filtering,
+    /// enrichment lookup, and managed-key formation. Google's OpenAI-compat
+    /// `/models` returns `models/`-prefixed ids while its chat endpoint (and
+    /// the models.dev snapshot) use the bare id — stripping canonicalizes to
+    /// the bare form. `None` = no stripping (the id is used verbatim). The
+    /// strip is a no-op when the prefix is absent, so it is safe even if a
+    /// listing returns some ids already bare.
+    strip_listing_id_prefix: Option<&'static str>,
 }

 const KIMI_CODE_SPEC: PlatformSpec = PlatformSpec {
@@ -163,6 +171,7 @@ const KIMI_CODE_SPEC: PlatformSpec = PlatformSpec {
    chat_compat: PlatformChatCompat::Kimi,
    key_header: PlatformKeyHeader::Bearer,
    restrict_to_enriched: false,
+    strip_listing_id_prefix: None,
 };

 const MOONSHOT_CN_SPEC: PlatformSpec = PlatformSpec {
@@ -185,6 +194,7 @@ const MOONSHOT_CN_SPEC: PlatformSpec = PlatformSpec {
    chat_compat: PlatformChatCompat::Kimi,
    key_header: PlatformKeyHeader::Bearer,
    restrict_to_enriched: false,
+    strip_listing_id_prefix: None,
 };

 const MOONSHOT_AI_SPEC: PlatformSpec = PlatformSpec {
@@ -207,6 +217,7 @@ const MOONSHOT_AI_SPEC: PlatformSpec = PlatformSpec {
    chat_compat: PlatformChatCompat::Kimi,
    key_header: PlatformKeyHeader::Bearer,
    restrict_to_enriched: false,
+    strip_listing_id_prefix: None,
 };

 /// Base-URL override for OpenAI (dev/test escape hatch).
@@ -234,6 +245,7 @@ const OPENAI_SPEC: PlatformSpec = PlatformSpec {
    chat_compat: PlatformChatCompat::Passthrough,
    key_header: PlatformKeyHeader::Bearer,
    restrict_to_enriched: true,
+    strip_listing_id_prefix: None,
 };

 /// Base-URL override for Anthropic (dev/test escape hatch).
@@ -261,6 +273,7 @@ const ANTHROPIC_SPEC: PlatformSpec = PlatformSpec {
    chat_compat: PlatformChatCompat::Passthrough,
    key_header: PlatformKeyHeader::XApiKey,
    restrict_to_enriched: false,
+    strip_listing_id_prefix: None,
 };

 /// Base-URL override for DeepSeek (dev/test escape hatch).
@@ -288,6 +301,7 @@ const DEEPSEEK_SPEC: PlatformSpec = PlatformSpec {
    chat_compat: PlatformChatCompat::DeepSeek,
    key_header: PlatformKeyHeader::Bearer,
    restrict_to_enriched: false,
+    strip_listing_id_prefix: None,
 };

 /// Base-URL override for Groq (dev/test escape hatch).
@@ -315,6 +329,7 @@ const GROQ_SPEC: PlatformSpec = PlatformSpec {
    // The listing carries whisper/tts entries; keep tool-calling chat
    // models only.
    restrict_to_enriched: true,
+    strip_listing_id_prefix: None,
 };

 /// Base-URL override for Mistral (dev/test escape hatch).
@@ -344,6 +359,7 @@ const MISTRAL_SPEC: PlatformSpec = PlatformSpec {
    // The listing carries embed/moderation/OCR entries; keep tool-calling
    // chat models only.
    restrict_to_enriched: true,
+    strip_listing_id_prefix: None,
 };

 /// Base-URL override for Fireworks (dev/test escape hatch).
@@ -372,6 +388,36 @@ const FIREWORKS_SPEC: PlatformSpec = PlatformSpec {
    // The inference /models listing can include embedding/non-chat models;
    // keep tool-calling enrichment-known models only.
    restrict_to_enriched: true,
+    strip_listing_id_prefix: None,
+};
+
+/// Base-URL override for Google Gemini (dev/test escape hatch).
+pub const GOOGLE_BASE_URL_ENV: &str = "KIGI_GOOGLE_BASE_URL";
+
+const GOOGLE_SPEC: PlatformSpec = PlatformSpec {
+    id: "google",
+    display_name: "Google Gemini",
+    base_url: BaseUrlSource::EnvOr {
+        env: GOOGLE_BASE_URL_ENV,
+        // Gemini's OpenAI-compatibility shim (bare model ids, Bearer key).
+        default: "https://generativelanguage.googleapis.com/v1beta/openai",
+    },
+    uses_oauth: false,
+    allowed_model_prefixes: None,
+    api_key_envs: &["GEMINI_API_KEY"],
+    vendor: "Google",
+    console_host: Some("aistudio.google.com"),
+    login_label: Some("Google Gemini (API key)"),
+    models_dev_id: Some("google"),
+    wire_serves_metadata: false,
+    wire_api: PlatformWireApi::ChatCompletions,
+    listing: ListingDialect::OpenAi,
+    chat_compat: PlatformChatCompat::Passthrough,
+    key_header: PlatformKeyHeader::Bearer,
+    // The compat listing carries embedding/tts/image models; keep
+    // tool-calling enrichment-known chat models only.
+    restrict_to_enriched: true,
+    strip_listing_id_prefix: Some("models/"),
 };

 /// The platform registry. Platforms are compiled-in spec rows; there is no
@@ -396,12 +442,14 @@ pub enum PlatformId {
    Mistral,
    /// Fireworks AI platform API (API key, OpenAI-compatible ChatCompletions).
    Fireworks,
+    /// Google Gemini platform API (API key, OpenAI-compatibility shim).
+    Google,
 }

 impl PlatformId {
    /// All platforms, in catalog precedence order: the subscription channel
    /// first so "default model = first list item" favors it when present.
-    pub const ALL: [PlatformId; 9] = [
+    pub const ALL: [PlatformId; 10] = [
        Self::KimiCode,
        Self::MoonshotCn,
        Self::MoonshotAi,
@@ -411,6 +459,7 @@ impl PlatformId {
        Self::Groq,
        Self::Mistral,
        Self::Fireworks,
+        Self::Google,
    ];

    /// The registry row backing this platform (single source of per-platform
@@ -426,6 +475,7 @@ impl PlatformId {
            Self::Groq => &GROQ_SPEC,
            Self::Mistral => &MISTRAL_SPEC,
            Self::Fireworks => &FIREWORKS_SPEC,
+            Self::Google => &GOOGLE_SPEC,
        }
    }

@@ -516,6 +566,12 @@ impl PlatformId {
        self.spec().restrict_to_enriched
    }

+    /// Prefix to strip from live-listing model ids before filter/enrich/key
+    /// (e.g. Google's `models/`). `None` = use the id verbatim.
+    pub fn strip_listing_id_prefix(self) -> Option<&'static str> {
+        self.spec().strip_listing_id_prefix
+    }
+
    /// Model-listing endpoint shape + headers.
    pub fn listing(self) -> ListingDialect {
        self.spec().listing
@@ -1060,6 +1116,27 @@ mod tests {
        );
    }

+    /// Google's compat listing returns `models/`-prefixed ids; the spec
+    /// must declare the strip so they canonicalize to the bare snapshot
+    /// form. Every other platform uses ids verbatim.
+    #[test]
+    fn only_google_strips_a_listing_id_prefix() {
+        assert_eq!(
+            PlatformId::Google.strip_listing_id_prefix(),
+            Some("models/")
+        );
+        for p in PlatformId::ALL {
+            if p != PlatformId::Google {
+                assert_eq!(
+                    p.strip_listing_id_prefix(),
+                    None,
+                    "{} must use listing ids verbatim",
+                    p.as_str()
+                );
+            }
+        }
+    }
+
    #[test]
    fn platform_ids_round_trip() {
        for p in PlatformId::ALL {
@@ -1085,9 +1162,10 @@ mod tests {
                PlatformId::Groq => 6,
                PlatformId::Mistral => 7,
                PlatformId::Fireworks => 8,
+                PlatformId::Google => 9,
            }
        }
-        const VARIANT_COUNT: usize = 9; // update together with `ordinal`
+        const VARIANT_COUNT: usize = 10; // update together with `ordinal`
        let mut seen: Vec<usize> = PlatformId::ALL.iter().map(|&p| ordinal(p)).collect();
        seen.sort_unstable();
        seen.dedup();
@@ -597,7 +597,8 @@ mod tests {
                "deepseek",
                "groq",
                "mistral",
-                "fireworks"
+                "fireworks",
+                "google"
            ]
        );
        assert_eq!(default_id(&built), Some(XAI_API_KEY_METHOD_ID));
@@ -629,7 +630,8 @@ mod tests {
                "deepseek",
                "groq",
                "mistral",
-                "fireworks"
+                "fireworks",
+                "google"
            ]
        );
        assert_eq!(default_id(&built), Some(CACHED_TOKEN_AUTH_METHOD_ID));
@@ -654,7 +656,8 @@ mod tests {
                "deepseek",
                "groq",
                "mistral",
-                "fireworks"
+                "fireworks",
+                "google"
            ]
        );
        assert_eq!(default_id(&built), Some(CACHED_TOKEN_AUTH_METHOD_ID));
@@ -682,7 +685,8 @@ mod tests {
                "deepseek",
                "groq",
                "mistral",
-                "fireworks"
+                "fireworks",
+                "google"
            ]
        );
        assert_eq!(default_id(&built), None);
@@ -309,6 +309,19 @@ fn fetch_one_platform_models(
            })?
        }
    };
+    // Canonicalize listing ids before filtering/enrichment/keying. Google's
+    // OpenAI-compat `/models` returns `models/`-prefixed ids while its chat
+    // endpoint and the models.dev snapshot use the bare id — without this the
+    // enrichment lookup misses and `restrict_to_enriched` would empty the
+    // catalog. No-op for platforms with no configured prefix.
+    let mut data = data;
+    if let Some(prefix) = platform.strip_listing_id_prefix() {
+        for wire in &mut data {
+            if let Some(bare) = wire.id.strip_prefix(prefix) {
+                wire.id = bare.to_string();
+            }
+        }
+    }
    let total = data.len();
    let mut filtered = kigi_models::filter_allowed_models(platform, data);
    if filtered.len() != total {
@@ -1395,6 +1408,103 @@ mod tests {
        );
    }

+    /// Google/Gemini-cycle e2e: the OpenAI-compat listing returns
+    /// `models/`-PREFIXED ids (Google's real shape), which the Google spec's
+    /// `strip_listing_id_prefix` canonicalizes to the bare form the models.dev
+    /// snapshot + chat endpoint use — WITHOUT the strip, restrict_to_enriched
+    /// would silently drop every Gemini model. Embedding pollution is
+    /// restricted away; Passthrough dialect; bare id on the wire.
+    #[tokio::test(flavor = "multi_thread")]
+    #[serial_test::serial]
+    async fn google_compat_listing_strips_prefix_restricts_and_maps_passthrough() {
+        let platform_server = wiremock::MockServer::start().await;
+        wiremock::Mock::given(wiremock::matchers::method("GET"))
+            .and(wiremock::matchers::path("/models"))
+            .and(wiremock::matchers::header("Authorization", "Bearer gk-1"))
+            .respond_with(wiremock::ResponseTemplate::new(200).set_body_json(
+                serde_json::json!({ "data": [
+                    // Real Gemini compat shape: `models/`-prefixed ids.
+                    { "id": "models/gemini-2.5-pro", "object": "model" },
+                    { "id": "models/gemini-embedding-001", "object": "model" }
+                ]}),
+            ))
+            .expect(1)
+            .mount(&platform_server)
+            .await;
+        let modelsdev_server = wiremock::MockServer::start().await;
+        wiremock::Mock::given(wiremock::matchers::method("GET"))
+            .and(wiremock::matchers::path("/api.json"))
+            .respond_with(wiremock::ResponseTemplate::new(200).set_body_json(
+                serde_json::json!({ "google": { "models": {
+                    "gemini-2.5-pro": {
+                        "limit": {"context": 1048576, "output": 65536},
+                        "tool_call": true
+                    },
+                    "gemini-embedding-001": { "limit": {"context": 2048} }
+                }}}),
+            ))
+            .expect(1)
+            .mount(&modelsdev_server)
+            .await;
+        let cache_dir = tempfile::tempdir().unwrap();
+        let _base = kigi_test_support::EnvGuard::set(
+            kigi_models::GOOGLE_BASE_URL_ENV,
+            platform_server.uri(),
+        );
+        let _mdev = kigi_test_support::EnvGuard::set(
+            crate::agent::enrichment_fetch::MODELS_DEV_URL_ENV,
+            format!("{}/api.json", modelsdev_server.uri()),
+        );
+        let _mdev_cache = kigi_test_support::EnvGuard::set(
+            crate::agent::enrichment_fetch::MODELS_DEV_CACHE_DIR_ENV,
+            cache_dir.path(),
+        );
+        let endpoints = crate::agent::config::EndpointsConfig::default();
+        let keys = crate::agent::models::PlatformApiKeys::test_single(
+            kigi_models::PlatformId::Google,
+            "gk-1",
+        );
+        let result = tokio::task::spawn_blocking(move || {
+            fetch_platform_models_blocking(&endpoints, None, &keys)
+        })
+        .await
+        .unwrap()
+        .expect("fetch must succeed");
+        assert_eq!(
+            result
+                .models
+                .iter()
+                .map(|m| m.id.as_deref().unwrap_or_default())
+                .collect::<Vec<_>>(),
+            vec!["google/gemini-2.5-pro"],
+            "prefix stripped → matches bare snapshot → kept as the bare \
+             managed key; embedding (not tool-calling) dropped"
+        );
+        let entry = &result.models[0];
+        assert_eq!(
+            entry.context_window.get(),
+            1_048_576,
+            "enrichment matched the bare id"
+        );
+        assert_eq!(entry.max_completion_tokens, Some(65_536));
+        assert_eq!(
+            entry.model, "gemini-2.5-pro",
+            "the BARE Gemini id rides the wire (chat rejects the models/ prefix)"
+        );
+        let model_entry = crate::agent::config::ModelEntry::from_config_entry(entry);
+        let creds = crate::agent::config::ResolvedCredentials {
+            api_key: Some("gk-1".into()),
+            base_url: entry.base_url.clone(),
+            auth_type: kigi_chat_state::AuthType::ApiKey,
+            auth_scheme: Default::default(),
+        };
+        let cfg = crate::agent::config::sampling_config_for_model(&model_entry, creds, None);
+        assert_eq!(
+            cfg.chat_compat,
+            kigi_sampling_types::ChatCompat::Passthrough
+        );
+    }
+
    #[test]
    fn get_env_keys_parses_strings_and_rejects_non_strings() {
        use crate::agent::config::EnvKeys;
@@ -6894,7 +6894,7 @@ pub(crate) mod tests {
    #[test]
    fn pending_menu_items_lists_interactive_methods_plus_quit() {
        let items = pending_menu_items(&fresh_user_auth_methods(), None);
-        assert_eq!(items.len(), 10, "9 login rows + Quit, got {items:?}");
+        assert_eq!(items.len(), 11, "10 login rows + Quit, got {items:?}");
        assert!(
            matches!(&items[0], PendingMenuItem::Login { label } if label == "Kimi Code (OAuth)"),
            "row 0 must be the OAuth login, got {:?}",
@@ -6957,7 +6957,14 @@ pub(crate) mod tests {
                label: "Fireworks AI (API key)".into(),
            }
        );
-        assert_eq!(items[9], PendingMenuItem::Quit);
+        assert_eq!(
+            items[9],
+            PendingMenuItem::ApiKey {
+                target: PlatformLogin(kigi_shell::models::PlatformId::Google),
+                label: "Google Gemini (API key)".into(),
+            }
+        );
+        assert_eq!(items[10], PendingMenuItem::Quit);
        // The non-interactive methods must never appear as rows.
        let byok = kigi_shell::agent::auth_method::build_auth_methods(
            kigi_shell::agent::auth_method::AuthMethodsBuildInputs {
@@ -6968,7 +6975,7 @@ pub(crate) mod tests {
        );
        assert_eq!(
            pending_menu_items(&byok.methods, None).len(),
-            10,
+            11,
            "xai.api_key / cached_token must not add rows"
        );
    }