Gemini support

.java-version

@@ -0,0 +1 @@
+temurin-21

docs/configuration.md

@@ -57,7 +57,7 @@ A `.eca/rules` folder from the workspace root containing `.mdc` files with the r
 
 `.eca/rules/talk_funny.mdc`
 ```markdown
---- 
+---
 description: Use when responding anything
 ---
 
@@ -70,7 +70,7 @@ A `$XDG_CONFIG_HOME/eca/rules` or `~/.config/eca/rules` folder containing `.mdc`
 
 `~/.config/eca/rules/talk_funny.mdc`
 ```markdown
---- 
+---
 description: Use when responding anything
 ---
 
@@ -103,6 +103,87 @@ For MCP servers configuration, use the `mcpServers` config, example:
 }
 ```
 
+## Google Gemini
+
+ECA supports Google's Gemini models through two main APIs: the **Gemini Developer API** and the **Vertex AI API**. Each has distinct configuration and authentication methods. Configuration keys are set at the root of the config file.
+
+### Gemini Developer API (API Key)
+
+This is the simplest way to get started. It uses an API key from Google AI Studio.
+
+-   **Authentication:** API Key
+-   **Endpoint:** `https://generativelanguage.googleapis.com/v1beta/`
+-   **Configuration:**
+
+    ```json
+    {
+      "geminiApiKey": "YOUR_GEMINI_API_KEY"
+    }
+    ```
+
+    Alternatively, you can set the `GEMINI_API_KEY` environment variable.
+
+### Vertex AI (API Key)
+
+For projects integrated with Google Cloud Platform (GCP), you can use an API key associated with your GCP project.
+
+-   **Authentication:** API Key (from GCP)
+-   **Endpoint:** `https://<LOCATION>-aiplatform.googleapis.com/v1/`
+-   **Configuration:**
+
+    ```json
+    {
+      "googleApiKey": "YOUR_GCP_API_KEY",
+      "googleProjectId": "your-gcp-project-id",
+      "googleProjectLocation": "your-gcp-region"
+    }
+    ```
+    You can also use the `GOOGLE_API_KEY`, `GOOGLE_PROJECT_ID`, and `GOOGLE_PROJECT_LOCATION` environment variables.
+
+### Vertex AI (Application Default Credentials - ADC)
+
+For a more secure and flexible setup, you can use Application Default Credentials (ADC). This method is used when no `googleApiKey` is provided.
+
+There are two common ways to provide these credentials:
+
+1.  **User ADC (for local development):**
+    Authenticate with the `gcloud` CLI. ECA will automatically pick up the credentials.
+    ```bash
+    gcloud auth application-default login
+    ```
+
+2.  **Service Account ADC (for automated environments):**
+    Set the `GOOGLE_APPLICATION_CREDENTIALS` environment variable to the absolute path of your service account JSON file.
+    ```bash
+    export GOOGLE_APPLICATION_CREDENTIALS="/path/to/your/service-account-file.json"
+    ```
+    Alternatively, you can set this path in your configuration file:
+    ```json
+    {
+      "googleApplicationCredentials": "/path/to/your/service-account-file.json"
+    }
+    ```
+
+In all ADC scenarios, you still need to provide the project ID and location:
+
+```json
+{
+  "googleProjectId": "your-gcp-project-id",
+  "googleProjectLocation": "your-gcp-region"
+}
+```
+You can also set the `GOOGLE_PROJECT_ID` and `GOOGLE_PROJECT_LOCATION` environment variables.
+
+**Summary of Configuration Options:**
+
+| Key                              | Environment Variable                 | Description                                                                  | Required For                               |
+| -------------------------------- | ------------------------------------ | ---------------------------------------------------------------------------- | ------------------------------------------ |
+| `geminiApiKey`                   | `GEMINI_API_KEY`                     | API key for the Gemini Developer API.                                        | Gemini Developer API                       |
+| `googleApiKey`                   | `GOOGLE_API_KEY`                     | API key for Vertex AI. If not provided, ADC will be used.                    | Vertex AI (API Key)                        |
+| `googleProjectId`                | `GOOGLE_PROJECT_ID`                  | Your Google Cloud project ID.                                                | Vertex AI (API Key), Vertex AI (ADC)       |
+| `googleProjectLocation`          | `GOOGLE_PROJECT_LOCATION`            | The GCP region for your project (e.g., `us-central1`).                       | Vertex AI (API Key), Vertex AI (ADC)       |
+
+
 ## Custom LLM providers
 
 It's possible to configure ECA to be aware of custom LLM providers if they follow a API schema similar to currently supported ones (openai, anthropic), example for a custom hosted litellm server:
@@ -195,6 +276,10 @@ Example:
 interface Config {
     openaiApiKey?: string;
     anthropicApiKey?: string;
+    geminiApiKey?: string;
+    googleApiKey?: string;
+    googleProjectId?: string;
+    googleProjectLocation?: string;
     rules: [{path: string;}];
     commands: [{path: string;}];
     systemPromptTemplateFile?: string;
@@ -211,7 +296,7 @@ interface Config {
     mcpServers: {[key: string]: {
         command: string;
         args?: string[];
-        disabled?: boolean; 
+        disabled?: boolean;
     }};
     customProviders: {[key: string]: {
         api: 'openai' | 'anthropic';
@@ -248,10 +333,14 @@ interface Config {
 {
   "openaiApiKey" : null,
   "anthropicApiKey" : null,
+  "geminiApiKey": null,
+  "googleApiKey": null,
+  "googleProjectId": null,
+  "googleProjectLocation": null,
   "rules" : [],
   "commands" : [],
   "nativeTools": {"filesystem": {"enabled": true},
-                  "shell": {"enabled": true, 
+                  "shell": {"enabled": true,
                             "excludeCommands": []}},
   "disabledTools": [],
   "toolCall": {

docs/models.md

@@ -4,11 +4,13 @@ The models capabilities and configurations are retrieved from [models.dev](https
 
 ## Built-in providers and capabilities
 
+
 | model     | tools (MCP) | reasoning / thinking | prompt caching | web_search |
 |-----------|-------------|----------------------|----------------|------------|
 | OpenAI    | √           | √                    | √              | √          |
 | Anthropic | √           | √                    | √              | √          |
 | Ollama    | √           | √                    | X              | X          |
+| Google    | X           | √                    | X              | X          |
 
 ### OpenAI
 
@@ -30,6 +32,9 @@ The models capabilities and configurations are retrieved from [models.dev](https
 
 - [any local ollama model](https://ollama.com/search)
 
+### Gemini
+
+- https://ai.google.dev/
 ### Custom models for built-in providers
 
 Just configure the model in your eca `models` config, for more details check its [configuration](./configuration.md#adding-models).

src/eca/config.clj

@@ -11,13 +11,37 @@
    [clojure.core.memoize :as memoize]
    [clojure.java.io :as io]
    [clojure.string :as string]
+   [eca.logger :as logger]
    [eca.shared :as shared]))
 
 (set! *warn-on-reflection* true)
 
 (def initial-config
-  {:openaiApiKey nil
+  {;; LLM providers authentication configuration
+   :openaiApiKey nil
+
    :anthropicApiKey nil
+
+   ;; Gemini can be authenticated using different methods - setting only one of them is enough
+   ;; but will depend on the provider configuration, IAM and other bits outof the scope of this project.
+   ;; api key for Gemini API
+   :geminiApiKey nil
+
+   ;; Vertex AI:
+   :googleProjectId nil
+   :googleProjectLocation nil
+   ;; auth with  API key, or leave blank to use ADC (Application Default Credentials)
+   :googleApiKey nil
+
+   ;; Ollama API
+   :ollama {:host "http://localhost"
+            :port 11434
+            :useTools true
+            :think true}
+
+   :customProviders {}
+
+   ;; all other settings
    :rules []
    :commands []
    :nativeTools {:filesystem {:enabled true}
@@ -35,13 +59,15 @@
             "claude-sonnet-4-20250514" {:extraPayload {:thinking {:type "enabled" :budget_tokens 2048}}}
             "claude-opus-4-1-20250805" {:extraPayload {:thinking {:type "enabled" :budget_tokens 2048}}}
             "claude-opus-4-20250514" {:extraPayload {:thinking {:type "enabled" :budget_tokens 2048}}}
-            "claude-3-5-haiku-20241022" {:extraPayload {:thinking {:type "enabled" :budget_tokens 2048}}}}
-   :ollama {:host "http://localhost"
-            :port 11434
-            :useTools true
-            :think true}
+            "claude-3-5-haiku-20241022" {:extraPayload {:thinking {:type "enabled" :budget_tokens 2048}}}
+
+            "gemni-2.0-flash" {}
+            "gemini-2.0-flash-pro" {}
+            "gemini-2.0-pro" {}}
+
+
    :chat {:welcomeMessage "Welcome to ECA!\n\nType '/' for commands\n\n"}
-   :customProviders {}
+
    :index {:ignoreFiles [{:type :gitignore}]}})
 
 (defn get-env [env] (System/getenv env))
@@ -68,6 +94,7 @@
                             (io/file (get-property "user.home") ".config"))
         config-file (io/file xdg-config-home "eca" "config.json")]
     (when (.exists config-file)
+      (logger/debug "[CONFIG]" (format "Loading global config from %s" config-file))
       (safe-read-json-string (slurp config-file)))))
 
 (def ^:private config-from-global-file (memoize/ttl config-from-global-file* :ttl/threshold ttl-cache-config-ms))
@@ -79,6 +106,7 @@
       final-config
       (let [config-file (io/file (shared/uri->filename uri) ".eca" "config.json")]
         (when (.exists config-file)
+          (logger/debug "[CONFIG]" (format "Loading project config from %s" config-file))
           (safe-read-json-string (slurp config-file))))))
    {}
    roots))

src/eca/llm_api.clj

@@ -3,6 +3,7 @@
    [clojure.string :as string]
    [eca.config :as config]
    [eca.llm-providers.anthropic :as llm-providers.anthropic]
+   [eca.llm-providers.google :as llm-providers.google]
    [eca.llm-providers.ollama :as llm-providers.ollama]
    [eca.llm-providers.openai :as llm-providers.openai]
    [eca.logger :as logger]))
@@ -50,6 +51,28 @@
   (or (config/get-env "OPENAI_API_URL")
       llm-providers.openai/base-url))
 
+;; Google Gemini auth
+(defn ^:private gemini-api-key [config]
+  (or (:geminiApiKey config)
+      (config/get-env "GEMINI_API_KEY")))
+
+(defn ^:private google-api-key [config]
+  (or (:googleApiKey config)
+      (config/get-env "GOOGLE_API_KEY")))
+
+(defn ^:private google-project-id [config]
+  (or (:googleProjectId config)
+      (config/get-env "GOOGLE_PROJECT_ID")))
+
+(defn ^:private google-project-location [config]
+  (or (:googleProjectLocation config)
+      (config/get-env "GOOGLE_PROJECT_LOCATION")))
+
+(defn ^:private google-any-auth? [config]
+  (or (gemini-api-key config)
+      (and (google-project-id config) (google-project-location config) (google-api-key config))
+      (and (google-project-id config) (google-project-location config))))
+
 (defn default-model
   "Returns the default LLM model checking this waterfall:
   - Any custom provider with defaultModel set
@@ -69,6 +92,8 @@
               [:api-key-found "claude-sonnet-4-20250514"])
             (when (openai-api-key config)
               [:api-key-found "gpt-5"])
+            (when (google-any-auth? config)
+              [:google-auth-found "gemini-2.5-pro"])
             (when-let [ollama-model (first (filter #(string/starts-with? % config/ollama-model-prefix) (keys (:models db))))]
               [:ollama-running ollama-model])
             [:default "claude-sonnet-4-20250514"])]
@@ -148,6 +173,25 @@
           :api-key (anthropic-api-key config)}
          callbacks)
 
+
+        (= "google" provider)
+        (llm-providers.google/completion!
+         {:model model
+          :instructions instructions
+          :user-messages user-messages
+          :max-output-tokens max-output-tokens
+          :reason? (and reason? (:reason? model-config))
+          :past-messages past-messages
+          :tools tools
+          :web-search web-search
+          ;; NOTE: no :api-url here, because Google provider figures it out
+          ;;       because it depends on how we're authenticated
+          :gemini-api-key (gemini-api-key config)
+          :google-api-key (google-api-key config)
+          :google-project-id (google-project-id config)
+          :google-project-location (google-project-location config)}
+         callbacks)
+
         (string/starts-with? model config/ollama-model-prefix)
         (llm-providers.ollama/completion!
          {:host (-> config :ollama :host)