Merge pull request #16 from itsflaid/codex/add-model-routing-config

Add command-based AI model routing

Author: itsflaid

PRD.md

@@ -251,7 +251,7 @@ MVP success means DevMap can:
 
 ## 9. Core Commands — MVP
 
-Four commands. No more, no less.
+Four core project commands plus one configuration command.
 
 > Nothing removed. Nothing added until MVP is shipped.
 
@@ -422,6 +422,21 @@ No issues found.
 
 ---
 
+### `devmap config model`
+
+Set a global Groq model override or restore automatic command-based routing.
+
+```bash
+devmap config model llama-3.1-8b-instant
+devmap config model openai/gpt-oss-120b
+devmap config model auto
+```
+
+The override applies to AI-powered commands. `auto` restores the defaults in
+the model routing table.
+
+---
+
 ## 10. Generated Files
 
 DevMap uses generated files to create reusable context for humans and AI agents.
@@ -538,9 +553,10 @@ ai/
 
 | Command | Model | Reason |
 |---|---|---|
-| `analyze`, `ask` | `openai/gpt-oss-20b` | Production model, fast and cost-efficient |
-| `analyze --deep` | `llama-3.3-70b-versatile` | Better reasoning for detailed explanations |
-| Fallback | `llama-3.3-70b-versatile` | Production fallback when primary model is unavailable |
+| `ask` | `llama-3.1-8b-instant` | Fast model for focused codebase questions |
+| `analyze` | `openai/gpt-oss-20b` | Balanced architecture interpretation |
+| `analyze --deep` | `openai/gpt-oss-120b` | Heavy cross-module reasoning |
+| Fallback | `openai/gpt-oss-20b` | Production fallback when a different primary model is unavailable |
 
 If a model becomes unavailable, DevMap gracefully falls back. No raw provider errors shown to users.
 

docs/architecture.md

@@ -491,13 +491,16 @@ MVP default model routing:
 
 | Command          | Model                     |
 | ---------------- | ------------------------- |
+| `ask`            | `llama-3.1-8b-instant`    |
 | `analyze`        | `openai/gpt-oss-20b`      |
-| `ask`            | `openai/gpt-oss-20b`      |
-| `analyze --deep` | `llama-3.3-70b-versatile` |
-| Fallback         | `llama-3.3-70b-versatile` |
+| `analyze --deep` | `openai/gpt-oss-120b`     |
+| Fallback         | `openai/gpt-oss-20b`      |
 
 If a model is unavailable, DevMap should fall back gracefully.
 Only Groq production models should be used as public defaults.
+
+Users can override automatic routing with `devmap config model <model>`.
+Running `devmap config model auto` restores the defaults above.
 
 Raw provider errors should not be shown directly to users.
 

docs/commands.md

@@ -6,14 +6,15 @@
 
 ## Overview
 
-DevMap MVP provides four core commands:
+DevMap MVP provides four core project commands and one configuration command:
 
 * `devmap init`
 * `devmap analyze`
 * `devmap ask`
-* `devmap doctor`
-
-No additional commands should be added until the MVP is shipped.
+* `devmap doctor`
+* `devmap config model`
+
+No additional product commands should be added until the MVP is shipped.
 
 Future commands are documented in:
 
@@ -366,7 +367,7 @@ Use existing snapshot or re-analyze first?
 
 ---
 
-## `devmap doctor`
+## `devmap doctor`
 
 Run diagnostics for DevMap setup.
 
@@ -435,11 +436,34 @@ Issues found:
 * Output must be readable
 * Errors must be actionable
 * Do not expose raw stack traces
-* Mention what command the user should run next
-
----
-
-## Global Flags
+* Mention what command the user should run next
+
+---
+
+## `devmap config model`
+
+Set a global model override for AI-powered commands.
+
+### Usage
+
+```bash
+devmap config model llama-3.1-8b-instant
+devmap config model openai/gpt-oss-120b
+devmap config model auto
+```
+
+`auto` restores command-based routing:
+
+* `ask` uses `llama-3.1-8b-instant`
+* `analyze` uses `openai/gpt-oss-20b`
+* `analyze --deep` uses `openai/gpt-oss-120b`
+
+The command preserves the configured provider and API key. DevMap must be
+initialized before changing the model.
+
+---
+
+## Global Flags
 
 Available globally:
 

docs/for-me-personal/PROGRESS.md

@@ -4,6 +4,16 @@ Terakhir diperbarui: 2026-06-14
 
 ## Update 2026-06-14
 
+### Model Routing And Config
+
+- Default `devmap ask` memakai `llama-3.1-8b-instant`.
+- Standard `devmap analyze` tetap memakai `openai/gpt-oss-20b`.
+- `devmap analyze --deep` memakai `openai/gpt-oss-120b`.
+- Fallback model memakai `openai/gpt-oss-20b`.
+- Command `devmap config model <model>` dapat menetapkan override global tanpa
+  mengubah provider atau API key.
+- `devmap config model auto` mengembalikan routing default per command.
+
 ### Context Builder Token Optimization
 
 - Pertanyaan navigasi English seperti `where` dan `find` sekarang memakai

docs/for-me-personal/TEST.md

@@ -44,6 +44,32 @@ Periksa `Relevant Files` dan prompt token usage. Query pertama seharusnya
 memprioritaskan production CLI source dan memakai context jauh lebih kecil
 daripada default lama lima file dengan maksimal 200 baris per file.
 
+## Model Routing And Override
+
+Focused automated test:
+
+```powershell
+pnpm --filter devmap exec tsx --test test/config-command.test.ts test/analyze-ai.test.ts test/ask-command.test.ts
+```
+
+Expected automatic routing:
+
+- `ask`: `llama-3.1-8b-instant`
+- `analyze`: `openai/gpt-oss-20b`
+- `analyze --deep`: `openai/gpt-oss-120b`
+- fallback: `openai/gpt-oss-20b`
+
+Manual override:
+
+```powershell
+pnpm dev:cli config model openai/gpt-oss-120b
+pnpm dev:cli doctor
+pnpm dev:cli config model auto
+```
+
+The first command should preserve the existing API key and provider. The last
+command should restore automatic command-based routing.
+
 ## Urutan Testing Yang Direkomendasikan
 
 Untuk development harian:

packages/cli/src/ai/groq.ts

@@ -13,10 +13,10 @@ const MAX_RATE_LIMIT_DELAY_MS = 10_000;
 const MAX_RATE_LIMIT_RETRIES = 3;
 
 export const DEFAULT_AI_MODELS = {
-  ask: "openai/gpt-oss-20b",
+  ask: "llama-3.1-8b-instant",
   analyze: "openai/gpt-oss-20b",
-  deepAnalyze: "llama-3.3-70b-versatile",
-  fallback: "llama-3.3-70b-versatile"
+  deepAnalyze: "openai/gpt-oss-120b",
+  fallback: "openai/gpt-oss-20b"
 } as const;
 
 export type GroqClientDependencies = {

packages/cli/src/commands/config.ts

@@ -0,0 +1,43 @@
+import {
+  readConfig,
+  writeConfig,
+  type DevmapConfig
+} from "../utils/config.js";
+import { output } from "../utils/output.js";
+
+export type ConfigDependencies = {
+  loadConfig?: () => Promise<DevmapConfig | null>;
+  persistConfig?: (config: DevmapConfig) => Promise<void>;
+};
+
+export async function configModelCommand(
+  model: string,
+  dependencies: ConfigDependencies = {}
+): Promise<void> {
+  const selectedModel = model.trim();
+  if (!selectedModel) {
+    output.error("Model name cannot be empty.");
+    return;
+  }
+
+  const loadConfig = dependencies.loadConfig ?? readConfig;
+  const persistConfig = dependencies.persistConfig ?? writeConfig;
+  const config = await loadConfig();
+
+  if (!config) {
+    output.error("DevMap is not configured yet.");
+    output.note("Run devmap init before changing the model.");
+    return;
+  }
+
+  await persistConfig({
+    ...config,
+    model: selectedModel
+  });
+
+  output.success(
+    selectedModel === "auto"
+      ? "Restored automatic command-based model routing."
+      : `Default model override set to ${selectedModel}.`
+  );
+}

packages/cli/src/index.ts

@@ -2,6 +2,7 @@
 import { Command } from "commander";
 import { analyzeCommand } from "./commands/analyze.js";
 import { askCommand } from "./commands/ask.js";
+import { configModelCommand } from "./commands/config.js";
 import { doctorCommand } from "./commands/doctor.js";
 import { initCommand } from "./commands/init.js";
 import { printHelp } from "./utils/help.js";
@@ -34,6 +35,14 @@ program
   .argument("<question...>", "question to ask")
   .action(askCommand);
 
+program
+  .command("config")
+  .description("Update DevMap configuration")
+  .command("model")
+  .description("Set a model override or restore automatic routing")
+  .argument("<model>", "Groq model ID or auto")
+  .action(configModelCommand);
+
 program
   .command("doctor")
   .description("Diagnose DevMap setup")

packages/cli/src/utils/help.ts

@@ -4,6 +4,7 @@ const commands = [
   ["init", "Initialize DevMap configuration"],
   ["analyze", "Analyze project structure"],
   ["ask <question>", "Ask about your codebase"],
+  ["config model", "Set model override or automatic routing"],
   ["doctor", "Diagnose DevMap setup"]
 ] as const;
 

packages/cli/test/analyze-ai.test.ts

@@ -3,6 +3,7 @@ import { mkdtemp, rm, writeFile } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 import test from "node:test";
+import { DEFAULT_AI_MODELS } from "../src/ai/groq.js";
 import type {
   AiClient,
   AiCompletionRequest,
@@ -124,6 +125,56 @@ test("analyze warns and continues when package.json is malformed", async () => {
   }
 });
 
+test("analyze auto routing uses 20B normally and 120B for deep analysis", async () => {
+  const projectRoot = await mkdtemp(join(tmpdir(), "devmap-model-routing-"));
+  const requests: AiCompletionRequest[] = [];
+  const client: AiClient = {
+    async complete(request): Promise<AiCompletionResult> {
+      requests.push(request);
+      return {
+        content: "Architecture result.",
+        model: request.model
+      };
+    }
+  };
+
+  try {
+    await writeFile(
+      join(projectRoot, "package.json"),
+      JSON.stringify({ name: "model-routing-fixture" }),
+      "utf8"
+    );
+    await writeFile(join(projectRoot, "index.ts"), "export const ready = true;\n", "utf8");
+
+    const dependencies = {
+      loadConfig: async () => ({
+        provider: "groq" as const,
+        apiKey: "gsk_fixture",
+        model: "auto"
+      }),
+      createAiClient: () => client
+    };
+
+    await captureOutput(() => analyzeCommand(
+      projectRoot,
+      { fresh: true },
+      dependencies
+    ));
+    await captureOutput(() => analyzeCommand(
+      projectRoot,
+      { deep: true, fresh: true },
+      dependencies
+    ));
+
+    assert.equal(requests[0]?.model, DEFAULT_AI_MODELS.analyze);
+    assert.equal(requests[1]?.model, DEFAULT_AI_MODELS.deepAnalyze);
+    assert.equal(requests[0]?.fallbackModel, DEFAULT_AI_MODELS.fallback);
+    assert.equal(requests[1]?.fallbackModel, DEFAULT_AI_MODELS.fallback);
+  } finally {
+    await rm(projectRoot, { recursive: true, force: true });
+  }
+});
+
 function stripAnsi(value: string): string {
   return value.replace(/\u001B\[[0-9;]*m/g, "");
 }