Files
stack/apps/api/src/ollama/ollama.service.ts
Jason Woltje 1e5fcd19a4 feat(#59): implement wiki-link parser
- Created wiki-link-parser.ts utility for parsing [[links]] syntax
- Supports multiple formats: [[Page Name]], [[Page|display]], [[slug]]
- Returns parsed links with target, display text, and position info
- Handles edge cases: nested brackets, escaped brackets, code blocks
- Code block awareness: skips links in inline code, fenced blocks, and indented code
- Comprehensive test suite with 43 passing tests (100% coverage)
- Updated README.md with parser documentation

Implements KNOW-007 (Issue #59) - Wiki-style linking foundation
2026-01-29 17:42:49 -06:00

345 lines
8.6 KiB
TypeScript

import { Injectable, Inject, HttpException, HttpStatus } from "@nestjs/common";
import type {
GenerateOptionsDto,
GenerateResponseDto,
ChatMessage,
ChatOptionsDto,
ChatResponseDto,
EmbedResponseDto,
ListModelsResponseDto,
HealthCheckResponseDto,
} from "./dto";
/**
* Configuration for Ollama service
*/
export interface OllamaConfig {
mode: "local" | "remote";
endpoint: string;
model: string;
timeout: number;
}
/**
* Service for interacting with Ollama API
* Supports both local and remote Ollama instances
*/
@Injectable()
export class OllamaService {
constructor(
@Inject("OLLAMA_CONFIG")
private readonly config: OllamaConfig
) {}
/**
* Generate text from a prompt
* @param prompt - The text prompt to generate from
* @param options - Generation options (temperature, max_tokens, etc.)
* @param model - Optional model override (defaults to config model)
* @returns Generated text response
*/
async generate(
prompt: string,
options?: GenerateOptionsDto,
model?: string
): Promise<GenerateResponseDto> {
const url = `${this.config.endpoint}/api/generate`;
const requestBody = {
model: model || this.config.model,
prompt,
stream: false,
...(options && {
options: this.mapGenerateOptions(options),
}),
};
try {
const controller = new AbortController();
const timeoutId = setTimeout(() => controller.abort(), this.config.timeout);
const response = await fetch(url, {
method: "POST",
headers: {
"Content-Type": "application/json",
},
body: JSON.stringify(requestBody),
signal: controller.signal,
});
clearTimeout(timeoutId);
if (!response.ok) {
throw new HttpException(
`Ollama API error: ${response.statusText}`,
response.status
);
}
const data = await response.json();
return data as GenerateResponseDto;
} catch (error: unknown) {
if (error instanceof HttpException) {
throw error;
}
const errorMessage =
error instanceof Error ? error.message : "Unknown error";
throw new HttpException(
`Failed to connect to Ollama: ${errorMessage}`,
HttpStatus.SERVICE_UNAVAILABLE
);
}
}
/**
* Complete a chat conversation
* @param messages - Array of chat messages
* @param options - Chat options (temperature, max_tokens, etc.)
* @param model - Optional model override (defaults to config model)
* @returns Chat completion response
*/
async chat(
messages: ChatMessage[],
options?: ChatOptionsDto,
model?: string
): Promise<ChatResponseDto> {
const url = `${this.config.endpoint}/api/chat`;
const requestBody = {
model: model || this.config.model,
messages,
stream: false,
...(options && {
options: this.mapChatOptions(options),
}),
};
try {
const controller = new AbortController();
const timeoutId = setTimeout(() => controller.abort(), this.config.timeout);
const response = await fetch(url, {
method: "POST",
headers: {
"Content-Type": "application/json",
},
body: JSON.stringify(requestBody),
signal: controller.signal,
});
clearTimeout(timeoutId);
if (!response.ok) {
throw new HttpException(
`Ollama API error: ${response.statusText}`,
response.status
);
}
const data = await response.json();
return data as ChatResponseDto;
} catch (error: unknown) {
if (error instanceof HttpException) {
throw error;
}
const errorMessage =
error instanceof Error ? error.message : "Unknown error";
throw new HttpException(
`Failed to connect to Ollama: ${errorMessage}`,
HttpStatus.SERVICE_UNAVAILABLE
);
}
}
/**
* Generate embeddings for text
* @param text - The text to generate embeddings for
* @param model - Optional model override (defaults to config model)
* @returns Embedding vector
*/
async embed(text: string, model?: string): Promise<EmbedResponseDto> {
const url = `${this.config.endpoint}/api/embeddings`;
const requestBody = {
model: model || this.config.model,
prompt: text,
};
try {
const controller = new AbortController();
const timeoutId = setTimeout(() => controller.abort(), this.config.timeout);
const response = await fetch(url, {
method: "POST",
headers: {
"Content-Type": "application/json",
},
body: JSON.stringify(requestBody),
signal: controller.signal,
});
clearTimeout(timeoutId);
if (!response.ok) {
throw new HttpException(
`Ollama API error: ${response.statusText}`,
response.status
);
}
const data = await response.json();
return data as EmbedResponseDto;
} catch (error: unknown) {
if (error instanceof HttpException) {
throw error;
}
const errorMessage =
error instanceof Error ? error.message : "Unknown error";
throw new HttpException(
`Failed to connect to Ollama: ${errorMessage}`,
HttpStatus.SERVICE_UNAVAILABLE
);
}
}
/**
* List available models
* @returns List of available Ollama models
*/
async listModels(): Promise<ListModelsResponseDto> {
const url = `${this.config.endpoint}/api/tags`;
try {
const controller = new AbortController();
const timeoutId = setTimeout(() => controller.abort(), this.config.timeout);
const response = await fetch(url, {
method: "GET",
signal: controller.signal,
});
clearTimeout(timeoutId);
if (!response.ok) {
throw new HttpException(
`Ollama API error: ${response.statusText}`,
response.status
);
}
const data = await response.json();
return data as ListModelsResponseDto;
} catch (error: unknown) {
if (error instanceof HttpException) {
throw error;
}
const errorMessage =
error instanceof Error ? error.message : "Unknown error";
throw new HttpException(
`Failed to connect to Ollama: ${errorMessage}`,
HttpStatus.SERVICE_UNAVAILABLE
);
}
}
/**
* Check health and connectivity of Ollama instance
* @returns Health check status
*/
async healthCheck(): Promise<HealthCheckResponseDto> {
try {
const controller = new AbortController();
const timeoutId = setTimeout(() => controller.abort(), 5000); // 5s timeout for health check
const response = await fetch(`${this.config.endpoint}/api/tags`, {
method: "GET",
signal: controller.signal,
});
clearTimeout(timeoutId);
if (response.ok) {
return {
status: "healthy",
mode: this.config.mode,
endpoint: this.config.endpoint,
available: true,
};
} else {
return {
status: "unhealthy",
mode: this.config.mode,
endpoint: this.config.endpoint,
available: false,
error: `HTTP ${response.status}: ${response.statusText}`,
};
}
} catch (error: unknown) {
const errorMessage =
error instanceof Error ? error.message : "Unknown error";
return {
status: "unhealthy",
mode: this.config.mode,
endpoint: this.config.endpoint,
available: false,
error: errorMessage,
};
}
}
/**
* Map GenerateOptionsDto to Ollama API options format
*/
private mapGenerateOptions(
options: GenerateOptionsDto
): Record<string, unknown> {
const mapped: Record<string, unknown> = {};
if (options.temperature !== undefined) {
mapped.temperature = options.temperature;
}
if (options.top_p !== undefined) {
mapped.top_p = options.top_p;
}
if (options.max_tokens !== undefined) {
mapped.num_predict = options.max_tokens;
}
if (options.stop !== undefined) {
mapped.stop = options.stop;
}
return mapped;
}
/**
* Map ChatOptionsDto to Ollama API options format
*/
private mapChatOptions(options: ChatOptionsDto): Record<string, unknown> {
const mapped: Record<string, unknown> = {};
if (options.temperature !== undefined) {
mapped.temperature = options.temperature;
}
if (options.top_p !== undefined) {
mapped.top_p = options.top_p;
}
if (options.max_tokens !== undefined) {
mapped.num_predict = options.max_tokens;
}
if (options.stop !== undefined) {
mapped.stop = options.stop;
}
return mapped;
}
}