feat(session): persist model tier overrides per session

Store per-session config in SQLite and route /model and /reset through command fast-paths so channel sessions keep independent model selection across reconnects and restarts.

src/backends/native/orchestrator.test.ts

@@ -4,7 +4,10 @@ import { ModelRouter } from '../../models/router.js';
 import type { ChatResponse, ModelClient } from '../../models/types.js';
 import { ToolRegistry, ToolExecutor } from '../../tools/index.js';
 import { HookEngine } from '../../hooks/engine.js';
-import type { SubAgentRequest } from './orchestrator.js';
+import { MemoryStore } from '../../memory/store.js';
+import { mkdtempSync, rmSync } from 'fs';
+import { tmpdir } from 'os';
+import { join } from 'path';
 
 describe('AgentOrchestrator', () => {
   let mockDefaultClient: ModelClient;
@@ -33,6 +36,14 @@ describe('AgentOrchestrator', () => {
     });
   });
 
+  const requireClient = (tier: 'default' | 'fast' | 'complex'): ModelClient => {
+    const client = mockRouter.getClient(tier);
+    if (!client) {
+      throw new Error(`Expected ${tier} model client to exist in test router`);
+    }
+    return client;
+  };
+
   describe('delegate()', () => {
     it('routes to the correct tier when specified', async () => {
       const orchestrator = new AgentOrchestrator({
@@ -69,7 +80,7 @@ describe('AgentOrchestrator', () => {
       });
       const mockToolExecutor = new ToolExecutor(mockToolRegistry, hooks);
 
-      const mockFastChatClient = mockRouter.getClient('fast')!;
+      const mockFastChatClient = requireClient('fast');
       const mockFastChatFn = vi.fn().mockResolvedValue({
         content: 'response with tools',
         stopReason: 'end_turn',
@@ -298,7 +309,7 @@ describe('AgentOrchestrator', () => {
 
   describe('process()', () => {
     it('proxies to NativeAgent for user messages', async () => {
-      const mockDefaultChatClient = mockRouter.getClient('default')!;
+      const mockDefaultChatClient = requireClient('default');
       const mockDefaultChatFn = vi.fn().mockResolvedValue({
         content: 'Agent response',
         stopReason: 'end_turn',
@@ -355,6 +366,88 @@ describe('AgentOrchestrator', () => {
       expect(history[4]).toEqual({ role: 'user', content: 'Tell me about yourself' });
       expect(history[5]).toEqual({ role: 'assistant', content: 'default response' });
     });
+
+    it('uses adaptive memory injection strategy when configured', async () => {
+      const tempDir = mkdtempSync(join(tmpdir(), 'flynn-orchestrator-memory-'));
+      const memoryStore = new MemoryStore({ dir: tempDir, maxContextTokens: 2000 });
+      memoryStore.writeCategory('user', 'preferences', 'User prefers concise output.', 'replace');
+
+      const mockDefaultChatClient = requireClient('default');
+      const mockDefaultChatFn = vi.fn().mockResolvedValue({
+        content: 'Agent response',
+        stopReason: 'end_turn',
+        usage: { inputTokens: 50, outputTokens: 25 },
+      } as ChatResponse);
+      Object.assign(mockDefaultChatClient, { chat: mockDefaultChatFn });
+
+      const orchestrator = new AgentOrchestrator({
+        modelRouter: mockRouter,
+        systemPrompt: 'You are a helpful agent.',
+        primaryTier: 'default',
+        delegation: {
+          compaction: 'fast',
+          memory_extraction: 'default',
+          classification: 'complex',
+          tool_summarisation: 'default',
+          complex_reasoning: 'complex',
+        },
+        maxDelegationDepth: 10,
+        memoryStore,
+        memoryInjectionStrategy: 'adaptive',
+        memoryMaxInjectionTokens: 100,
+      });
+
+      await orchestrator.process('Keep this concise please');
+
+      expect(mockDefaultChatFn).toHaveBeenCalled();
+      const callArgs = mockDefaultChatFn.mock.calls[0][0];
+      expect(callArgs.system).toContain('# Memory Context');
+      expect(callArgs.system).toContain('concise');
+
+      rmSync(tempDir, { recursive: true, force: true });
+    });
+
+    it('falls back to default memory context when adaptive injection errors', async () => {
+      const tempDir = mkdtempSync(join(tmpdir(), 'flynn-orchestrator-memory-fallback-'));
+      const memoryStore = new MemoryStore({ dir: tempDir, maxContextTokens: 2000 });
+      memoryStore.write('user', 'Fallback memory content', 'replace');
+      const getPromptSectionsSpy = vi.spyOn(memoryStore, 'getPromptSections').mockImplementationOnce(() => {
+        throw new Error('boom');
+      });
+
+      const mockDefaultChatClient = requireClient('default');
+      const mockDefaultChatFn = vi.fn().mockResolvedValue({
+        content: 'Agent response',
+        stopReason: 'end_turn',
+        usage: { inputTokens: 50, outputTokens: 25 },
+      } as ChatResponse);
+      Object.assign(mockDefaultChatClient, { chat: mockDefaultChatFn });
+
+      const orchestrator = new AgentOrchestrator({
+        modelRouter: mockRouter,
+        systemPrompt: 'You are a helpful agent.',
+        primaryTier: 'default',
+        delegation: {
+          compaction: 'fast',
+          memory_extraction: 'default',
+          classification: 'complex',
+          tool_summarisation: 'default',
+          complex_reasoning: 'complex',
+        },
+        maxDelegationDepth: 10,
+        memoryStore,
+        memoryInjectionStrategy: 'adaptive',
+        memoryMaxInjectionTokens: 100,
+      });
+
+      await orchestrator.process('test message');
+
+      const callArgs = mockDefaultChatFn.mock.calls[0][0];
+      expect(callArgs.system).toContain('Fallback memory content');
+
+      getPromptSectionsSpy.mockRestore();
+      rmSync(tempDir, { recursive: true, force: true });
+    });
   });
 
   describe('reset()', () => {