feat: add llama.cpp provider support

README.md

@@ -90,6 +90,25 @@ export OLLAMA_HOST=0.0.0.0
 
 This will make Ollama listen on all interfaces, including IPv6 and IPv4, resolving the connection issue. You can add this line to your shell configuration file (like `.bashrc` or `.zshrc`) to make it persistent across sessions.
 
+### Running locally with llama.cpp
+
+You can run OpenCommit with local models served by [llama.cpp](https://github.com/ggerganov/llama.cpp):
+
+- install and build llama.cpp
+- start the server with a GGUF model: `./llama-server -m model.gguf --port 8080`
+- run (in your project directory):
+
+```sh
+git add <files...>
+oco config set OCO_AI_PROVIDER='llamacpp' OCO_API_URL='http://localhost:8080'
+```
+
+If the server is running on a different machine, set the URL accordingly:
+
+```sh
+oco config set OCO_API_URL='http://192.168.1.10:8080'
+```
+
 ### Flags
 
 There are multiple optional flags that can be used with the `oco` command:
@@ -122,7 +141,7 @@ Create a `.env` file and add OpenCommit config variables there like this:
 
 ```env
 ...
-OCO_AI_PROVIDER=<openai (default), anthropic, azure, ollama, gemini, flowise, deepseek, aimlapi>
+OCO_AI_PROVIDER=<openai (default), anthropic, azure, ollama, llamacpp, gemini, flowise, deepseek, aimlapi>
 OCO_API_KEY=<your OpenAI API token> // or other LLM provider API token
 OCO_API_URL=<may be used to set proxy path to OpenAI api>
 OCO_API_CUSTOM_HEADERS=<JSON string of custom HTTP headers to include in API requests>
@@ -227,14 +246,16 @@ oco models --provider anthropic
 
 By default OpenCommit uses [OpenAI](https://openai.com).
 
-You could switch to [Azure OpenAI Service](https://learn.microsoft.com/azure/cognitive-services/openai/) or Flowise or Ollama.
+You could switch to [Azure OpenAI Service](https://learn.microsoft.com/azure/cognitive-services/openai/) or Flowise or Ollama or llama.cpp.
 
 ```sh
 oco config set OCO_AI_PROVIDER=azure OCO_API_KEY=<your_azure_api_key> OCO_API_URL=<your_azure_endpoint>
 
 oco config set OCO_AI_PROVIDER=flowise OCO_API_KEY=<your_flowise_api_key> OCO_API_URL=<your_flowise_endpoint>
 
 oco config set OCO_AI_PROVIDER=ollama OCO_API_KEY=<your_ollama_api_key> OCO_API_URL=<your_ollama_endpoint>
+
+oco config set OCO_AI_PROVIDER=llamacpp OCO_API_URL=<your_llamacpp_endpoint>
 ```
 
 ### Use with Proxy

src/commands/config.ts

@@ -579,6 +579,8 @@ const getDefaultModel = (provider: string | undefined): string => {
   switch (provider) {
     case 'ollama':
       return '';
+    case 'llamacpp':
+      return '';
     case 'mlx':
       return '';
     case 'anthropic':
@@ -797,8 +799,10 @@ export const configValidators = {
         'deepseek',
         'aimlapi',
         'openrouter'
-      ].includes(value) || value.startsWith('ollama'),
-      `${value} is not supported yet, use 'ollama', 'mlx', 'anthropic', 'azure', 'gemini', 'flowise', 'mistral', 'deepseek', 'aimlapi' or 'openai' (default)`
+      ].includes(value) ||
+        value.startsWith('ollama') ||
+        value.startsWith('llamacpp'),
+      `${value} is not supported yet, use 'ollama', 'llamacpp', 'mlx', 'anthropic', 'azure', 'gemini', 'flowise', 'mistral', 'deepseek', 'aimlapi' or 'openai' (default)`
     );
 
     return value;
@@ -854,6 +858,7 @@ export const configValidators = {
 
 export enum OCO_AI_PROVIDER_ENUM {
   OLLAMA = 'ollama',
+  LLAMACPP = 'llamacpp',
   OPENAI = 'openai',
   ANTHROPIC = 'anthropic',
   GEMINI = 'gemini',
@@ -880,6 +885,7 @@ export const PROVIDER_API_KEY_URLS: Record<string, string | null> = {
   [OCO_AI_PROVIDER_ENUM.AIMLAPI]: 'https://aimlapi.com/app/keys',
   [OCO_AI_PROVIDER_ENUM.AZURE]: 'https://portal.azure.com/',
   [OCO_AI_PROVIDER_ENUM.OLLAMA]: null,
+  [OCO_AI_PROVIDER_ENUM.LLAMACPP]: null,
   [OCO_AI_PROVIDER_ENUM.MLX]: null,
   [OCO_AI_PROVIDER_ENUM.FLOWISE]: null,
   [OCO_AI_PROVIDER_ENUM.TEST]: null

src/commands/setup.ts

@@ -24,6 +24,7 @@ const PROVIDER_DISPLAY_NAMES: Record<string, string> = {
   [OCO_AI_PROVIDER_ENUM.OPENAI]: 'OpenAI (GPT-4o, GPT-4)',
   [OCO_AI_PROVIDER_ENUM.ANTHROPIC]: 'Anthropic (Claude Sonnet, Opus)',
   [OCO_AI_PROVIDER_ENUM.OLLAMA]: 'Ollama (Free, runs locally)',
+  [OCO_AI_PROVIDER_ENUM.LLAMACPP]: 'llama.cpp (Free, runs locally)',
   [OCO_AI_PROVIDER_ENUM.GEMINI]: 'Google Gemini',
   [OCO_AI_PROVIDER_ENUM.GROQ]: 'Groq (Fast inference, free tier)',
   [OCO_AI_PROVIDER_ENUM.MISTRAL]: 'Mistral AI',
@@ -37,7 +38,8 @@ const PROVIDER_DISPLAY_NAMES: Record<string, string> = {
 const PRIMARY_PROVIDERS = [
   OCO_AI_PROVIDER_ENUM.OPENAI,
   OCO_AI_PROVIDER_ENUM.ANTHROPIC,
-  OCO_AI_PROVIDER_ENUM.OLLAMA
+  OCO_AI_PROVIDER_ENUM.OLLAMA,
+  OCO_AI_PROVIDER_ENUM.LLAMACPP
 ];
 
 const OTHER_PROVIDERS = [
@@ -53,13 +55,15 @@ const OTHER_PROVIDERS = [
 
 const NO_API_KEY_PROVIDERS = [
   OCO_AI_PROVIDER_ENUM.OLLAMA,
+  OCO_AI_PROVIDER_ENUM.LLAMACPP,
   OCO_AI_PROVIDER_ENUM.MLX,
   OCO_AI_PROVIDER_ENUM.TEST
 ];
 
 const MODEL_REQUIRED_PROVIDERS = [
   OCO_AI_PROVIDER_ENUM.OLLAMA,
-  OCO_AI_PROVIDER_ENUM.MLX
+  OCO_AI_PROVIDER_ENUM.MLX,
+  OCO_AI_PROVIDER_ENUM.LLAMACPP
 ];
 
 async function selectProvider(): Promise<string | symbol> {
@@ -335,6 +339,37 @@ async function setupOllama(): Promise<{
   };
 }
 
+async function setupLlamaCpp(): Promise<{
+  provider: string;
+  model: string;
+  apiUrl: string;
+} | null> {
+  console.log(chalk.cyan('\n  llama.cpp - Free Local AI\n'));
+  console.log(chalk.dim('  Setup steps:'));
+  console.log(
+    chalk.dim('  1. Install: https://github.com/ggerganov/llama.cpp')
+  );
+  console.log(
+    chalk.dim('  2. Start server: llama-server -m <model.gguf> --port 8080\n')
+  );
+
+  const defaultUrl = 'http://localhost:8080';
+
+  const apiUrl = await text({
+    message: 'llama.cpp server URL (press Enter for default):',
+    placeholder: defaultUrl,
+    defaultValue: defaultUrl
+  });
+
+  if (isCancel(apiUrl)) return null;
+
+  return {
+    provider: OCO_AI_PROVIDER_ENUM.LLAMACPP,
+    model: '',
+    apiUrl: (apiUrl as string) || defaultUrl
+  };
+}
+
 export async function runSetup(): Promise<boolean> {
   intro(chalk.bgCyan(' Welcome to OpenCommit! '));
 
@@ -382,6 +417,19 @@ export async function runSetup(): Promise<boolean> {
       OCO_MODEL: model,
       OCO_API_KEY: 'mlx' // Placeholder
     };
+  } else if (provider === OCO_AI_PROVIDER_ENUM.LLAMACPP) {
+    const llamacppConfig = await setupLlamaCpp();
+    if (!llamacppConfig) {
+      outro('Setup cancelled');
+      return false;
+    }
+
+    config = {
+      OCO_AI_PROVIDER: llamacppConfig.provider,
+      OCO_MODEL: llamacppConfig.model,
+      OCO_API_URL: llamacppConfig.apiUrl,
+      OCO_API_KEY: 'llamacpp' // Placeholder
+    };
   } else {
     // Standard provider flow: API key then model
     const apiKey = await getApiKey(provider as string);

src/engine/llamacpp.ts

@@ -0,0 +1,52 @@
+import axios, { AxiosInstance } from 'axios';
+import { OpenAI } from 'openai';
+import { normalizeEngineError } from '../utils/engineErrorHandler';
+import { removeContentTags } from '../utils/removeContentTags';
+import { AiEngine, AiEngineConfig } from './Engine';
+
+interface LlamaCppConfig extends AiEngineConfig {}
+
+const DEFAULT_LLAMACPP_URL = 'http://localhost:8080';
+const LLAMACPP_CHAT_PATH = '/v1/chat/completions';
+
+export class LlamaCppEngine implements AiEngine {
+  config: LlamaCppConfig;
+  client: AxiosInstance;
+  private chatUrl: string;
+
+  constructor(config: LlamaCppConfig) {
+    this.config = config;
+
+    const baseUrl = config.baseURL || DEFAULT_LLAMACPP_URL;
+    this.chatUrl = `${baseUrl}${LLAMACPP_CHAT_PATH}`;
+
+    const headers = {
+      'Content-Type': 'application/json',
+      ...config.customHeaders
+    };
+
+    this.client = axios.create({ headers });
+  }
+
+  async generateCommitMessage(
+    messages: Array<OpenAI.Chat.Completions.ChatCompletionMessageParam>
+  ): Promise<string | undefined> {
+    const params: Record<string, any> = {
+      model: this.config.model ?? '',
+      messages,
+      temperature: 0,
+      top_p: 0.1,
+      repeat_penalty: 1.1,
+      stream: false
+    };
+    try {
+      const response = await this.client.post(this.chatUrl, params);
+
+      const choices = response.data.choices;
+      const message = choices[0].message;
+      return removeContentTags(message?.content, 'think');
+    } catch (error) {
+      throw normalizeEngineError(error, 'llamacpp', this.config.model);
+    }
+  }
+}

src/utils/engine.ts

@@ -4,6 +4,7 @@ import { AzureEngine } from '../engine/azure';
 import { AiEngine } from '../engine/Engine';
 import { FlowiseEngine } from '../engine/flowise';
 import { GeminiEngine } from '../engine/gemini';
+import { LlamaCppEngine } from '../engine/llamacpp';
 import { OllamaEngine } from '../engine/ollama';
 import { OpenAiEngine } from '../engine/openAi';
 import { MistralAiEngine } from '../engine/mistral';
@@ -40,6 +41,9 @@ export function getEngine(): AiEngine {
         ollamaThink: config.OCO_OLLAMA_THINK
       });
 
+    case OCO_AI_PROVIDER_ENUM.LLAMACPP:
+      return new LlamaCppEngine(DEFAULT_CONFIG);
+
     case OCO_AI_PROVIDER_ENUM.ANTHROPIC:
       return new AnthropicEngine(DEFAULT_CONFIG);
 

src/utils/modelCache.ts

@@ -87,6 +87,23 @@ export async function fetchOllamaModels(
   }
 }
 
+export async function fetchLlamaCppModels(
+  baseUrl: string = 'http://localhost:8080'
+): Promise<string[]> {
+  try {
+    const response = await fetch(`${baseUrl}/v1/models`);
+
+    if (!response.ok) {
+      return [];
+    }
+
+    const data = await response.json();
+    return data.data?.map((m: { id: string }) => m.id) || [];
+  } catch {
+    return [];
+  }
+}
+
 export async function fetchAnthropicModels(apiKey: string): Promise<string[]> {
   try {
     const response = await fetch('https://api.anthropic.com/v1/models', {
@@ -231,6 +248,10 @@ export async function fetchModelsForProvider(
       models = await fetchOllamaModels(baseUrl);
       break;
 
+    case OCO_AI_PROVIDER_ENUM.LLAMACPP:
+      models = await fetchLlamaCppModels(baseUrl);
+      break;
+
     case OCO_AI_PROVIDER_ENUM.ANTHROPIC:
       if (apiKey) {
         models = await fetchAnthropicModels(apiKey);

test/unit/llamacpp.test.ts

@@ -0,0 +1,103 @@
+import { LlamaCppEngine } from '../../src/engine/llamacpp';
+
+describe('LlamaCppEngine', () => {
+  it('sends request to /v1/chat/completions', async () => {
+    const engine = new LlamaCppEngine({
+      apiKey: 'llamacpp',
+      model: 'llama-3',
+      maxTokensOutput: 500,
+      maxTokensInput: 4096
+    });
+
+    const post = jest.fn().mockResolvedValue({
+      data: {
+        choices: [
+          {
+            message: {
+              content: 'feat: add support for llama.cpp provider'
+            }
+          }
+        ]
+      }
+    });
+
+    engine.client = { post } as any;
+
+    await engine.generateCommitMessage([
+      { role: 'user', content: 'diff --git a/file b/file' }
+    ]);
+
+    expect(post).toHaveBeenCalledWith(
+      'http://localhost:8080/v1/chat/completions',
+      expect.objectContaining({
+        temperature: 0,
+        top_p: 0.1,
+        repeat_penalty: 1.1,
+        stream: false
+      })
+    );
+  });
+
+  it('uses custom baseURL when provided', async () => {
+    const engine = new LlamaCppEngine({
+      apiKey: 'llamacpp',
+      model: 'llama-3',
+      maxTokensOutput: 500,
+      maxTokensInput: 4096,
+      baseURL: 'http://192.168.1.10:8080'
+    });
+
+    const post = jest.fn().mockResolvedValue({
+      data: {
+        choices: [
+          {
+            message: {
+              content: 'fix: resolve connection issue'
+            }
+          }
+        ]
+      }
+    });
+
+    engine.client = { post } as any;
+
+    await engine.generateCommitMessage([
+      { role: 'user', content: 'diff --git a/file b/file' }
+    ]);
+
+    expect(post).toHaveBeenCalledWith(
+      'http://192.168.1.10:8080/v1/chat/completions',
+      expect.anything()
+    );
+  });
+
+  it('strips <think> tags from response content', async () => {
+    const engine = new LlamaCppEngine({
+      apiKey: 'llamacpp',
+      model: 'llama-3',
+      maxTokensOutput: 500,
+      maxTokensInput: 4096
+    });
+
+    const post = jest.fn().mockResolvedValue({
+      data: {
+        choices: [
+          {
+            message: {
+              content:
+                '<think>reasoning here</think>feat: add support for llama.cpp provider'
+            }
+          }
+        ]
+      }
+    });
+
+    engine.client = { post } as any;
+
+    const result = await engine.generateCommitMessage([
+      { role: 'user', content: 'diff --git a/file b/file' }
+    ]);
+
+    expect(result).toBe('feat: add support for llama.cpp provider');
+  });
+});