yaml-flow 3.0.0 → 3.1.1

package/examples/inference/pluggable-adapters.ts

@@ -0,0 +1,254 @@
+/**
+ * Inference Example: Pluggable LLM Adapters
+ *
+ * Shows how to create adapters for different LLM providers.
+ * Each adapter implements InferenceAdapter.analyze(prompt) → string.
+ *
+ * Run with: npx tsx examples/inference/pluggable-adapters.ts
+ */
+
+import {
+  createLiveGraph,
+  schedule,
+} from '../../src/continuous-event-graph/index.js';
+import {
+  inferCompletions,
+  applyInferences,
+  buildInferencePrompt,
+  createCliAdapter,
+  createHttpAdapter,
+} from '../../src/inference/index.js';
+import type { InferenceAdapter } from '../../src/inference/index.js';
+import type { GraphConfig } from '../../src/continuous-event-graph/types.js';
+
+// ============================================================================
+// 1. Adapter Examples (uncomment the one for your LLM provider)
+// ============================================================================
+
+/**
+ * OpenAI Adapter
+ * npm install openai
+ */
+function createOpenAIAdapter(apiKey: string, model = 'gpt-4o'): InferenceAdapter {
+  return {
+    analyze: async (prompt: string) => {
+      // import OpenAI from 'openai';
+      // const client = new OpenAI({ apiKey });
+      // const response = await client.chat.completions.create({
+      //   model,
+      //   messages: [{ role: 'user', content: prompt }],
+      //   temperature: 0.1,  // low temperature for deterministic analysis
+      // });
+      // return response.choices[0].message.content ?? '[]';
+
+      // Mock for demo
+      return '[]';
+    },
+  };
+}
+
+/**
+ * Azure OpenAI Adapter
+ * npm install openai
+ */
+function createAzureOpenAIAdapter(
+  endpoint: string,
+  apiKey: string,
+  deployment: string,
+): InferenceAdapter {
+  return {
+    analyze: async (prompt: string) => {
+      // import OpenAI from 'openai';
+      // const client = new OpenAI({
+      //   apiKey,
+      //   baseURL: `${endpoint}/openai/deployments/${deployment}`,
+      //   defaultQuery: { 'api-version': '2024-02-15-preview' },
+      //   defaultHeaders: { 'api-key': apiKey },
+      // });
+      // const response = await client.chat.completions.create({
+      //   model: deployment,
+      //   messages: [{ role: 'user', content: prompt }],
+      //   temperature: 0.1,
+      // });
+      // return response.choices[0].message.content ?? '[]';
+
+      return '[]';
+    },
+  };
+}
+
+/**
+ * Anthropic (Claude) Adapter
+ * npm install @anthropic-ai/sdk
+ */
+function createAnthropicAdapter(apiKey: string, model = 'claude-sonnet-4-20250514'): InferenceAdapter {
+  return {
+    analyze: async (prompt: string) => {
+      // import Anthropic from '@anthropic-ai/sdk';
+      // const client = new Anthropic({ apiKey });
+      // const response = await client.messages.create({
+      //   model,
+      //   max_tokens: 1024,
+      //   messages: [{ role: 'user', content: prompt }],
+      // });
+      // return response.content[0].type === 'text' ? response.content[0].text : '[]';
+
+      return '[]';
+    },
+  };
+}
+
+/**
+ * Local/Custom Adapter — call any HTTP endpoint
+ */
+function createCustomAdapter(url: string, headers: Record<string, string> = {}): InferenceAdapter {
+  return {
+    analyze: async (prompt: string) => {
+      const response = await fetch(url, {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json', ...headers },
+        body: JSON.stringify({ prompt }),
+      });
+      const data = await response.json() as { response?: string };
+      return data.response ?? '[]';
+    },
+  };
+}
+
+// ============================================================================
+// 1b. Built-in Adapter Factories (no boilerplate needed)
+// ============================================================================
+
+/**
+ * Ollama via CLI — runs models locally
+ * Install: https://ollama.com
+ */
+const ollamaCliAdapter = createCliAdapter({
+  command: 'ollama',
+  args: (prompt) => ['run', 'llama3', prompt],
+});
+
+/**
+ * Ollama via HTTP — same model, HTTP interface
+ * Start with: ollama serve
+ */
+const ollamaHttpAdapter = createHttpAdapter({
+  url: 'http://localhost:11434/api/generate',
+  buildBody: (prompt) => ({ model: 'llama3', prompt, stream: false }),
+  extractResponse: (json) => json.response as string,
+});
+
+/**
+ * llm CLI — Simon Willison's LLM tool
+ * Install: pip install llm
+ */
+const llmCliAdapter = createCliAdapter({
+  command: 'llm',
+  args: () => ['--no-stream'],
+  stdin: true, // pipe prompt via stdin (better for long prompts)
+});
+
+/**
+ * GitHub Copilot CLI — use gh copilot for inference
+ * Install: gh extension install github/gh-copilot
+ */
+const ghCopilotAdapter = createCliAdapter({
+  command: 'gh',
+  args: (prompt) => ['copilot', 'suggest', '-t', 'shell', prompt],
+  timeout: 30_000,
+});
+
+/**
+ * Custom script adapter — run your own wrapper script
+ * The script receives the prompt as argument (or stdin) and prints JSON to stdout
+ */
+const customScriptAdapter = createCliAdapter({
+  command: 'python',
+  args: (prompt) => ['scripts/infer.py', '--json', prompt],
+  cwd: '/path/to/project',
+  env: { MODEL: 'gpt-4o', TEMPERATURE: '0.1' },
+  timeout: 60_000,
+});
+
+// ============================================================================
+// 2. Demo: Use the prompt builder standalone
+// ============================================================================
+
+const config: GraphConfig = {
+  settings: { completion: 'all-tasks' },
+  tasks: {
+    'code-reviewed': {
+      provides: ['review-done'],
+      description: 'Code review completed by team lead',
+      inference: {
+        criteria: 'PR approved with at least 2 approvals and no open comments',
+        keywords: ['pull-request', 'code-review', 'approval'],
+        suggestedChecks: ['check PR status for "approved"', 'verify no open comments'],
+        autoDetectable: true,
+      },
+    },
+    'tests-passed': {
+      requires: ['review-done'],
+      provides: ['tests-green'],
+      description: 'CI pipeline tests all passing',
+      inference: {
+        criteria: 'All CI checks green, code coverage above 80%',
+        keywords: ['ci', 'tests', 'coverage', 'pipeline'],
+        suggestedChecks: ['check CI status', 'verify coverage report'],
+        autoDetectable: true,
+      },
+    },
+    'deployed-staging': {
+      requires: ['tests-green'],
+      provides: ['staging-live'],
+      description: 'Deployed to staging environment',
+      inference: {
+        criteria: 'Staging URL returns HTTP 200 with correct version',
+        keywords: ['staging', 'deployment', 'version'],
+        autoDetectable: true,
+      },
+    },
+  },
+};
+
+async function main() {
+  const live = createLiveGraph(config);
+
+  // Just build the prompt — useful for debugging or using with your own LLM code
+  const prompt = buildInferencePrompt(live, {
+    context: 'PR #142: 3 approvals, 0 open comments. CI: all 47 tests pass, coverage 84%.',
+  });
+
+  console.log('=== Generated LLM Prompt ===');
+  console.log(prompt);
+  console.log('\n=== End Prompt ===');
+  console.log(`\nPrompt length: ${prompt.length} characters`);
+  console.log('This prompt can be sent to any LLM provider.');
+
+  // Demo: use a mock adapter to show the full flow
+  const mockAdapter: InferenceAdapter = {
+    analyze: async () => JSON.stringify([
+      { taskName: 'code-reviewed', confidence: 0.97, reasoning: '3 approvals, 0 open comments — PR is clearly approved.' },
+      { taskName: 'tests-passed', confidence: 0.92, reasoning: '47/47 tests pass, coverage 84% exceeds 80% threshold.' },
+    ]),
+  };
+
+  const result = await inferCompletions(live, mockAdapter, {
+    context: 'PR #142: 3 approvals, 0 open comments. CI: all 47 tests pass, coverage 84%.',
+  });
+
+  console.log('\n=== Inference Results ===');
+  for (const s of result.suggestions) {
+    console.log(`  ${s.taskName}: ${(s.confidence * 100).toFixed(0)}% — ${s.reasoning}`);
+  }
+
+  // Apply above threshold
+  const updated = applyInferences(live, result, 0.9);
+  console.log('\n=== After Applying (threshold 90%) ===');
+  for (const [name, state] of Object.entries(updated.state.tasks)) {
+    console.log(`  ${name}: ${state.status}`);
+  }
+  console.log('  Eligible:', schedule(updated).eligible);
+}
+
+main().catch(console.error);