import { ollamaRequest, ollamaRequestInference, ollamaRequestPull } from "../ollama/client.js";
import {
OllamaGenerateResponseSchema,
OllamaChatResponseSchema,
OllamaListResponseSchema,
OllamaShowResponseSchema,
OllamaEmbeddingsResponseSchema,
OllamaPullResponseSchema,
} from "../ollama/types.js";
import {
OllamaGenerateSchema,
OllamaChatSchema,
OllamaShowSchema,
OllamaPullSchema,
OllamaEmbeddingsSchema,
type OllamaGenerateArgs,
type OllamaChatArgs,
type OllamaShowArgs,
type OllamaPullArgs,
type OllamaEmbeddingsArgs,
} from "./schemas.js";
import { assertValidModelName } from "./validators.js";
import { processImages } from "../utils/images.js";
/**
* Generate text using an Ollama model
*/
export async function handleOllamaGenerate(raw: unknown): Promise<string> {
const args: OllamaGenerateArgs = OllamaGenerateSchema.parse(raw);
assertValidModelName(args.model);
// Process images if provided (convert file paths to base64)
const images = await processImages(args.images);
const data = await ollamaRequestInference(
"/api/generate",
{
method: "POST",
body: JSON.stringify({
model: args.model,
prompt: args.prompt,
system: args.system,
images,
stream: false,
options: {
temperature: args.temperature,
num_predict: args.max_tokens,
},
}),
},
OllamaGenerateResponseSchema
);
return data.response;
}
/**
* Have a multi-turn chat conversation with an Ollama model
*/
export async function handleOllamaChat(raw: unknown): Promise<string> {
const args: OllamaChatArgs = OllamaChatSchema.parse(raw);
assertValidModelName(args.model);
// Process images in messages (convert file paths to base64)
const messagesWithImages = await Promise.all(
args.messages.map(async (msg) => ({
role: msg.role,
content: msg.content,
images: await processImages(msg.images),
}))
);
const data = await ollamaRequestInference(
"/api/chat",
{
method: "POST",
body: JSON.stringify({
model: args.model,
messages: messagesWithImages,
stream: false,
options: {
temperature: args.temperature,
},
}),
},
OllamaChatResponseSchema
);
return data.message.content;
}
/**
* List all locally available Ollama models
*/
export async function handleOllamaList(): Promise<string> {
const data = await ollamaRequest(
"/api/tags",
{},
undefined,
OllamaListResponseSchema
);
if (!data.models || data.models.length === 0) {
return "No models found. Use ollama_pull to download a model.";
}
const modelList = data.models.map((m) => {
const sizeGB = (m.size / (1024 * 1024 * 1024)).toFixed(2);
return `- ${m.name} (${sizeGB} GB)`;
});
return `Available models:\n${modelList.join("\n")}`;
}
/**
* Show detailed information about a specific model
*/
export async function handleOllamaShow(raw: unknown): Promise<string> {
const args: OllamaShowArgs = OllamaShowSchema.parse(raw);
assertValidModelName(args.model);
const data = await ollamaRequest(
"/api/show",
{
method: "POST",
body: JSON.stringify({ name: args.model }),
},
undefined,
OllamaShowResponseSchema
);
return JSON.stringify(data, null, 2);
}
/**
* Pull (download) a model from the Ollama library
*/
export async function handleOllamaPull(raw: unknown): Promise<string> {
const args: OllamaPullArgs = OllamaPullSchema.parse(raw);
assertValidModelName(args.model);
// Use extended timeout for model pull operations
await ollamaRequestPull(
"/api/pull",
{
method: "POST",
body: JSON.stringify({
name: args.model,
stream: false,
}),
},
OllamaPullResponseSchema
);
return `Successfully pulled model: ${args.model}`;
}
/**
* Generate embeddings for text using an Ollama model
*/
export async function handleOllamaEmbeddings(raw: unknown): Promise<string> {
const args: OllamaEmbeddingsArgs = OllamaEmbeddingsSchema.parse(raw);
assertValidModelName(args.model);
const data = await ollamaRequest(
"/api/embed",
{
method: "POST",
body: JSON.stringify({
model: args.model,
input: args.input,
}),
},
undefined,
OllamaEmbeddingsResponseSchema
);
return JSON.stringify(
{
model: args.model,
dimensions: data.embeddings[0]?.length || 0,
embeddings: data.embeddings,
},
null,
2
);
}
