Add OLLAMA as Model Provider

core/.env.example

@@ -28,6 +28,20 @@ XAI_API_KEY=
 XAI_MODEL=
 
 
+#Set to Use for New OLLAMA provider 
+OLLAMA_SERVER_URL=   #Leave blank for default localhost:11434
+OLLAMA_MODEL= 
+OLLAMA_EMBEDDING_MODEL=     #default mxbai-embed-large
+#To use custom model types for different tasks set these
+SMALL_OLLAMA_MODEL=         #default llama3.2
+MEDIUM_OLLAMA_MODEL=        #default herems3
+LARGE_OLLAMA_MODEL=         #default hermes3:70b
+
+#to still use the original LOCALLLAMA provider but with ollama
+LOCAL_LLAMA_PROVIDER= #"OLLAMA" #Leave blank for LLAMA-CPP or add OLLAMA
+
+
+
 # For asking Claude stuff
 ANTHROPIC_API_KEY=
 

core/package.json

@@ -81,6 +81,7 @@
         "@ai-sdk/google-vertex": "^0.0.42",
         "@ai-sdk/groq": "^0.0.3",
         "@ai-sdk/openai": "^0.0.70",
+        "ollama-ai-provider": "^0.16.1",
         "@anthropic-ai/sdk": "^0.30.1",
         "@cliqz/adblocker-playwright": "1.34.0",
         "@coral-xyz/anchor": "^0.30.1",

core/src/core/embedding.ts

@@ -10,8 +10,9 @@ export async function embed(runtime: IAgentRuntime, input: string) {
     // get the charcter, and handle by model type
     const model = models[runtime.character.settings.model];
 
-    if (model !== ModelProvider.OPENAI) {
+    if (model !== ModelProvider.OPENAI && model !== ModelProvider.OLLAMA) {
         return await runtime.llamaService.getEmbeddingResponse(input);
+        //ollama supports embedding api so just use that
     }
 
     const embeddingModel = models[runtime.modelProvider].model.embedding;
@@ -26,7 +27,8 @@ export async function embed(runtime: IAgentRuntime, input: string) {
         method: "POST",
         headers: {
             "Content-Type": "application/json",
-            Authorization: `Bearer ${runtime.token}`,
+            //Authorization: `Bearer ${runtime.token}`,
+            ...(runtime.modelProvider !== ModelProvider.OLLAMA && { Authorization: `Bearer ${runtime.token}` }),
         },
         body: JSON.stringify({
             input,
@@ -36,7 +38,8 @@ export async function embed(runtime: IAgentRuntime, input: string) {
     };
     try {
         const response = await fetch(
-            `${runtime.serverUrl}/embeddings`,
+            //`${runtime.serverUrl}/embeddings`,
+            `${runtime.serverUrl}${runtime.modelProvider === ModelProvider.OLLAMA ? '/v1' : ''}/embeddings`,
             requestOptions
         );
 

core/src/core/generation.ts

@@ -16,6 +16,8 @@ import { generateText as aiGenerateText } from "ai";
 import { createAnthropic } from "@ai-sdk/anthropic";
 import { prettyConsole } from "../index.ts";
 
+import { createOllama } from 'ollama-ai-provider';
+
 /**
  * Send a message to the model for a text generateText - receive a string back and parse how you'd like
  * @param opts - The options for the generateText request.
@@ -190,6 +192,30 @@ export async function generateText({
                 break;
             }
 
+            case ModelProvider.OLLAMA: {
+                console.log("Initializing Ollama model.");
+
+                const ollamaProvider = createOllama({
+                    baseURL: models[provider].endpoint + "/api",
+                })
+                const ollama = ollamaProvider(model);
+
+                console.log('****** MODEL\n', model)
+
+                const { text: ollamaResponse } = await aiGenerateText({
+                    model: ollama,
+                    prompt: context,
+                    temperature: temperature,
+                    maxTokens: max_response_length,
+                    frequencyPenalty: frequency_penalty,
+                    presencePenalty: presence_penalty,
+                });
+
+                response = ollamaResponse;
+            }
+                console.log("Received response from Ollama model.");
+                break;
+
             default: {
                 const errorMessage = `Unsupported provider: ${provider}`;
                 prettyConsole.error(errorMessage);

core/src/core/models.ts

@@ -10,6 +10,7 @@ type Models = {
     [ModelProvider.GOOGLE]: Model;
     [ModelProvider.CLAUDE_VERTEX]: Model;
     [ModelProvider.REDPILL]: Model;
+    [ModelProvider.OLLAMA]: Model;
     // TODO: add OpenRouter - feel free to do this :)
 };
 
@@ -169,6 +170,23 @@ const models: Models = {
             [ModelClass.EMBEDDING]: "text-embedding-3-small",
         },
     },
+    [ModelProvider.OLLAMA]: {
+        settings: {
+            stop: [],
+            maxInputTokens: 128000,
+            maxOutputTokens: 8192,
+            frequency_penalty: 0.0,
+            presence_penalty: 0.0,
+            temperature: 0.3,
+        },
+        endpoint: process.env.OLLAMA_SERVER_URL || "http://localhost:11434", 
+        model: {
+            [ModelClass.SMALL]: process.env.SMALL_OLLAMA_MODEL || process.env.OLLAMA_MODEL || "llama3.2",
+            [ModelClass.MEDIUM]: process.env.MEDIUM_OLLAMA_MODEL ||process.env.OLLAMA_MODEL || "hermes3",
+            [ModelClass.LARGE]: process.env.LARGE_OLLAMA_MODEL || process.env.OLLAMA_MODEL || "hermes3:70b",
+            [ModelClass.EMBEDDING]: process.env.OLLAMA_EMBEDDING_MODEL || "mxbai-embed-large"
+        },
+    },
 };
 
 export function getModel(provider: ModelProvider, type: ModelClass) {

core/src/core/types.ts

@@ -107,7 +107,8 @@ export enum ModelProvider {
     LLAMALOCAL = "llama_local",
     GOOGLE = "google",
     CLAUDE_VERTEX = "claude_vertex",
-    REDPILL = "redpill"
+    REDPILL = "redpill",
+    OLLAMA = "ollama"
 }
 
 /**