From 955aeed6ce8b4cea092d6947f20b3de6f32528bd Mon Sep 17 00:00:00 2001 From: mawkone Date: Sat, 16 May 2026 15:00:17 -0700 Subject: [PATCH] fix(ai): hardcode all default LLM references to gemini-3.1-pro-preview across monorepo --- vibn-agent-runner/src/llm.ts | 6 +++--- vibn-frontend/lib/ai/gemini-client.ts | 2 +- vibn-frontend/lib/ai/plan-extract.ts | 4 ++-- 3 files changed, 6 insertions(+), 6 deletions(-) diff --git a/vibn-agent-runner/src/llm.ts b/vibn-agent-runner/src/llm.ts index 0edd99d..ea3c172 100644 --- a/vibn-agent-runner/src/llm.ts +++ b/vibn-agent-runner/src/llm.ts @@ -12,7 +12,7 @@ import { v4 as uuidv4 } from "uuid"; // AnthropicVertexClient — Claude models via Anthropic Messages API on Vertex (us-east5) // // Model tier defaults (overridable via TIER_A/B/C_MODEL env vars): -// Tier A: gemini-2.5-flash — routing, summaries (API key, high quota) +// Tier A: gemini-3.1-pro-preview — routing, summaries (API key, high quota) // Tier B: claude-sonnet-4-6 — coding, feature work (Anthropic Vertex, us-east5) // Tier C: claude-sonnet-4-6 — complex decisions // ============================================================================= @@ -232,7 +232,7 @@ export class GeminiClient implements LLMClient { modelId: string; private temperature: number; - constructor(modelId = "gemini-2.5-flash", opts?: { temperature?: number }) { + constructor(modelId = "gemini-3.1-pro-preview", opts?: { temperature?: number }) { this.modelId = modelId; this.temperature = opts?.temperature ?? 0.2; } @@ -515,7 +515,7 @@ export class AnthropicVertexClient implements LLMClient { export type ModelTier = "A" | "B" | "C"; const TIER_MODELS: Record = { - A: process.env.TIER_A_MODEL ?? "gemini-2.5-flash", + A: process.env.TIER_A_MODEL ?? "gemini-3.1-pro-preview", B: process.env.TIER_B_MODEL ?? "claude-sonnet-4-6", C: process.env.TIER_C_MODEL ?? "claude-sonnet-4-6", }; diff --git a/vibn-frontend/lib/ai/gemini-client.ts b/vibn-frontend/lib/ai/gemini-client.ts index 43563d4..f43dff4 100644 --- a/vibn-frontend/lib/ai/gemini-client.ts +++ b/vibn-frontend/lib/ai/gemini-client.ts @@ -1,7 +1,7 @@ import type { LlmClient, StructuredCallArgs } from '@/lib/ai/llm-client'; import { zodToJsonSchema } from 'zod-to-json-schema'; -const DEFAULT_MODEL = process.env.GEMINI_MODEL || 'gemini-2.0-flash-exp'; +const DEFAULT_MODEL = process.env.GEMINI_MODEL || 'gemini-3.1-pro-preview'; const GOOGLE_API_KEY = process.env.GOOGLE_API_KEY || ''; const GEMINI_BASE_URL = 'https://generativelanguage.googleapis.com/v1beta/models'; diff --git a/vibn-frontend/lib/ai/plan-extract.ts b/vibn-frontend/lib/ai/plan-extract.ts index 9ef9079..cc5bfb9 100644 --- a/vibn-frontend/lib/ai/plan-extract.ts +++ b/vibn-frontend/lib/ai/plan-extract.ts @@ -6,14 +6,14 @@ * and auto-persist them via the same `fs_projects.data->plan` path used by * the Plan tab MCP tools. * - * The cheap model is configured via VIBN_CHEAP_MODEL (default: gemini-2.0-flash-001). + * The cheap model is configured via VIBN_CHEAP_MODEL (default: gemini-3.1-pro-preview). */ import { query } from "@/lib/db-postgres"; const GEMINI_API_KEY = process.env.GOOGLE_API_KEY || ""; const CHEAP_MODEL = - process.env.VIBN_CHEAP_MODEL || "gemini-2.0-flash-001"; + process.env.VIBN_CHEAP_MODEL || "gemini-3.1-pro-preview"; const GEMINI_BASE_URL = "https://generativelanguage.googleapis.com/v1beta"; interface PlanExtraction {