v2.0.5: FAST_MODEL routing + tool-use summaries + Qwen dispatch + Arena

Source-level recon of QwenLM/qwen-code (Apache-2.0) informed 4 lifts:

1. FAST_MODEL config: optional env var routes cheap LLM calls (titles,
   summaries, labeling) to a smaller model on llama-swap. auto_name.ts
   uses ctx.config.FAST_MODEL ?? session.model. Set FAST_MODEL=nemotron-
   nano-4b to avoid loading the 35B model for 20-token title generation.

2. Tool-use summaries (services/inference/tool-summaries.ts): utility
   that generates "git-commit-subject-style" labels for tool batches via
   a fast-model LLM call. System prompt + truncation logic ported from
   Qwen Code's toolUseSummary.ts. Exported via @boocode/server/inference
   for BooCoder's dispatcher to call after task completion.

3. Qwen as dispatchable agent: added to agent-probe.ts KNOWN_AGENTS.
   PTY dispatch builds: qwen -p "<task>" --output-format stream-json
   (NDJSON structured events over stdout). Env: OPENAI_BASE_URL +
   OPENAI_API_KEY points Qwen Code at llama-swap. execution_path CHECK
   constraint extended with 'qwen'.

4. Arena routes (routes/arena.ts): POST /api/arena dispatches the same
   task to N contestants (2-5, each with different agent/model), each
   getting its own task row linked by arena_id UUID. GET /api/arena/:id
   shows all contestants. POST /api/arena/:id/select/:task_id marks
   winner. Schema: arena_id column added to tasks.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

apps/coder/src/config.ts

@@ -23,6 +23,8 @@ const ConfigSchema = z.object({
   GITEA_TOKEN: z.string().optional(),
   GITEA_SSH_HOST: z.string().default('100.114.205.53:2222'),
   MCP_CONFIG_PATH: z.string().optional(),
+  // v2.0.5: cheaper model for titles, summaries, labeling.
+  FAST_MODEL: z.string().optional(),
   // SSH access to the host for external agent dispatch (Phase 5)
   BOOCODER_SSH_HOST: z.string().default('100.114.205.53'),
   BOOCODER_SSH_USER: z.string().default('samkintop'),

apps/coder/src/index.ts

@@ -27,6 +27,7 @@ import { registerPendingRoutes } from './routes/pending.js';
 import { registerTaskRoutes } from './routes/tasks.js';
 import { registerInboxRoutes } from './routes/inbox.js';
 import { registerStatsRoutes } from './routes/stats.js';
+import { registerArenaRoutes } from './routes/arena.js';
 import { registerWebSocket } from './routes/ws.js';
 // Phase 4: dispatcher + agent probe
 import { createDispatcher } from './services/dispatcher.js';
@@ -143,6 +144,7 @@ async function main() {
   registerTaskRoutes(app, sql, inferenceApi);
   registerInboxRoutes(app, sql);
   registerStatsRoutes(app, sql);
+  registerArenaRoutes(app, sql);
   registerWebSocket(app, sql, broker);
 
   // Serve static frontend (built web app). In production, the dist/ is

apps/coder/src/routes/arena.ts

@@ -0,0 +1,122 @@
+/**
+ * v2.0.5: Arena routes — competitive dispatch of the same task to multiple agents.
+ *
+ * POST /api/arena        — create an arena with 2-5 contestants
+ * GET  /api/arena/:id    — get all tasks in an arena
+ * POST /api/arena/:id/select/:task_id — mark a task as the arena winner
+ */
+import type { FastifyInstance } from 'fastify';
+import { z } from 'zod';
+import type { Sql } from '../db.js';
+
+const ContestantSchema = z.object({
+  agent: z.string().max(100).optional(),
+  model: z.string().max(200).optional(),
+});
+
+const CreateArenaBody = z.object({
+  project_id: z.string().uuid(),
+  input: z.string().min(1).max(64_000),
+  contestants: z.array(ContestantSchema).min(2).max(5),
+});
+
+interface TaskRow {
+  id: string;
+  agent: string | null;
+  model: string | null;
+  state: string;
+}
+
+export function registerArenaRoutes(app: FastifyInstance, sql: Sql): void {
+  // POST /api/arena — create a new arena
+  app.post('/api/arena', async (req, reply) => {
+    const parsed = CreateArenaBody.safeParse(req.body);
+    if (!parsed.success) {
+      reply.code(400);
+      return { error: 'invalid body', details: parsed.error.flatten() };
+    }
+
+    const { project_id, input, contestants } = parsed.data;
+    const arenaId = crypto.randomUUID();
+
+    const tasks: TaskRow[] = [];
+    for (const contestant of contestants) {
+      const [task] = await sql<TaskRow[]>`
+        INSERT INTO tasks (project_id, input, agent, model, arena_id)
+        VALUES (${project_id}, ${input}, ${contestant.agent ?? null}, ${contestant.model ?? null}, ${arenaId})
+        RETURNING id, agent, model, state
+      `;
+      tasks.push(task!);
+    }
+
+    reply.code(201);
+    return {
+      arena_id: arenaId,
+      tasks: tasks.map(t => ({
+        id: t.id,
+        agent: t.agent,
+        model: t.model,
+        state: t.state,
+      })),
+    };
+  });
+
+  // GET /api/arena/:arena_id — list all tasks in an arena
+  app.get<{ Params: { arena_id: string } }>('/api/arena/:arena_id', async (req, reply) => {
+    const { arena_id } = req.params;
+
+    // Validate UUID format
+    const uuidRegex = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i;
+    if (!uuidRegex.test(arena_id)) {
+      reply.code(400);
+      return { error: 'invalid arena_id format' };
+    }
+
+    const tasks = await sql`
+      SELECT id, project_id, state, input, output_summary, agent, model, execution_path, session_id, started_at, ended_at, created_at, arena_id
+      FROM tasks
+      WHERE arena_id = ${arena_id}
+      ORDER BY created_at
+    `;
+
+    if (tasks.length === 0) {
+      reply.code(404);
+      return { error: 'arena not found' };
+    }
+
+    return { arena_id, tasks };
+  });
+
+  // POST /api/arena/:arena_id/select/:task_id — mark the winner
+  app.post<{ Params: { arena_id: string; task_id: string } }>(
+    '/api/arena/:arena_id/select/:task_id',
+    async (req, reply) => {
+      const { arena_id, task_id } = req.params;
+
+      // Verify the task belongs to this arena
+      const rows = await sql<{ id: string; state: string; arena_id: string | null }[]>`
+        SELECT id, state, arena_id FROM tasks WHERE id = ${task_id}
+      `;
+
+      if (rows.length === 0) {
+        reply.code(404);
+        return { error: 'task not found' };
+      }
+
+      const task = rows[0]!;
+      if (task.arena_id !== arena_id) {
+        reply.code(409);
+        return { error: 'task does not belong to this arena' };
+      }
+
+      // Mark as selected via output_summary prefix (lightweight — no schema change)
+      await sql`
+        UPDATE tasks
+        SET output_summary = COALESCE('[SELECTED] ' || output_summary, '[SELECTED]')
+        WHERE id = ${task_id}
+      `;
+
+      return { selected: true, task_id, arena_id };
+    }
+  );
+}

apps/coder/src/schema.sql

@@ -31,7 +31,7 @@ CREATE TABLE IF NOT EXISTS tasks (
   ended_at TIMESTAMPTZ,
   created_at TIMESTAMPTZ NOT NULL DEFAULT clock_timestamp(),
   CONSTRAINT tasks_state_chk CHECK (state IN ('pending', 'running', 'completed', 'failed', 'blocked', 'cancelled')),
-  CONSTRAINT tasks_execution_path_chk CHECK (execution_path IS NULL OR execution_path IN ('native', 'acp', 'pty'))
+  CONSTRAINT tasks_execution_path_chk CHECK (execution_path IS NULL OR execution_path IN ('native', 'acp', 'pty', 'qwen'))
 );
 
 CREATE TABLE IF NOT EXISTS available_agents (
@@ -46,6 +46,18 @@ CREATE TABLE IF NOT EXISTS available_agents (
 -- v2.0.0 Phase 4: link tasks to their inference sessions.
 ALTER TABLE tasks ADD COLUMN IF NOT EXISTS session_id UUID REFERENCES sessions(id);
 
+-- v2.0.5: add 'qwen' to execution_path CHECK + arena_id column.
+ALTER TABLE tasks DROP CONSTRAINT IF EXISTS tasks_execution_path_chk;
+DO $$ BEGIN
+  IF NOT EXISTS (SELECT 1 FROM pg_constraint WHERE conname = 'tasks_execution_path_chk') THEN
+    ALTER TABLE tasks ADD CONSTRAINT tasks_execution_path_chk
+      CHECK (execution_path IS NULL OR execution_path IN ('native', 'acp', 'pty', 'qwen'));
+  END IF;
+END $$;
+
+-- v2.0.5: arena support — group tasks into competitive arenas.
+ALTER TABLE tasks ADD COLUMN IF NOT EXISTS arena_id UUID;
+
 -- Human inbox: tasks needing attention
 CREATE OR REPLACE VIEW human_inbox AS
   SELECT * FROM tasks WHERE state IN ('blocked', 'failed');

apps/coder/src/services/agent-probe.ts

@@ -7,6 +7,7 @@ const KNOWN_AGENTS: Array<{ name: string; supportsAcp: boolean }> = [
   { name: 'goose', supportsAcp: true },
   { name: 'claude', supportsAcp: false },
   { name: 'pi', supportsAcp: false },
+  { name: 'qwen', supportsAcp: false },
 ];
 
 /**

apps/coder/src/services/pty-dispatch.ts

@@ -8,6 +8,7 @@
  * Supported agents:
  * - claude: `claude -p --model <model>` (print mode, reads task from stdin)
  * - opencode: `echo <task> | opencode` (stdin pipe — exact flags TBD)
+ * - qwen: `qwen -p <task> --output-format stream-json` (NDJSON structured output)
  * - goose: stub (not yet supported)
  * - pi: stub (not yet supported)
  */
@@ -51,6 +52,12 @@ function buildAgentCommand(agent: string, task: string, model?: string): string
         ? `echo '${escapedTask}' | opencode --model '${model}'`
         : `echo '${escapedTask}' | opencode`;
 
+    case 'qwen':
+      // Qwen Code: structured JSON output mode for parseable events
+      return model
+        ? `qwen -p '${escapedTask}' --model '${model}' --output-format stream-json`
+        : `qwen -p '${escapedTask}' --output-format stream-json`;
+
     case 'goose':
       // Not yet verified for non-interactive use
       return null;

apps/server/src/config.ts

@@ -22,6 +22,9 @@ const ConfigSchema = z.object({
   // v1.15.0-mcp-multi: path to the MCP config JSON file. Default /data/mcp.json
   // (bind-mounted alongside AGENTS.md). File missing = no MCP (opt-in).
   MCP_CONFIG_PATH: z.string().optional(),
+  // v2.0.5: cheaper model for titles, summaries, labeling. Falls back to
+  // session model (auto_name) or DEFAULT_MODEL when unset.
+  FAST_MODEL: z.string().optional(),
 });
 
 export type Config = z.infer<typeof ConfigSchema>;

apps/server/src/services/auto_name.ts

@@ -67,7 +67,8 @@ export async function maybeAutoNameChat(
   const sessionRows = await ctx.sql<{ model: string }[]>`
     SELECT model FROM sessions WHERE id = ${sessionId}
   `;
-  const model = sessionRows[0]?.model;
+  // v2.0.5: prefer FAST_MODEL for cheap LLM calls (titles, summaries).
+  const model = ctx.config.FAST_MODEL ?? sessionRows[0]?.model;
   if (!model) return;
 
   const assistantMsg = await ctx.sql<{ content: string }[]>`

apps/server/src/services/inference/index.ts

@@ -20,3 +20,5 @@ export type {
 export type { ToolPhaseResult } from './tool-phase.js';
 export { detectDoomLoop, DOOM_LOOP_THRESHOLD } from './sentinels.js';
 export { buildMessagesPayload } from './payload.js';
+export { generateToolUseSummary } from './tool-summaries.js';
+export type { ToolInfo } from './tool-summaries.js';

apps/server/src/services/inference/tool-summaries.ts

@@ -0,0 +1,81 @@
+/**
+ * v2.0.5: Tool-use summary generation.
+ *
+ * After a batch of tool calls completes, fire a cheap LLM call to generate
+ * a "git-commit-subject-style" one-liner label describing what the tools
+ * accomplished. Ported from the Qwen Code source recon.
+ */
+import type { FastifyBaseLogger } from 'fastify';
+
+const TOOL_SUMMARY_SYSTEM_PROMPT = `Write a short summary label describing what these tool calls accomplished. Think git-commit-subject, not sentence. Past tense, most distinctive noun. Max 30 characters. Output ONLY the label.
+
+Examples:
+- Searched in auth/
+- Fixed NPE in UserService
+- Created signup endpoint
+- Read config.json
+- Ran failing tests`;
+
+const INPUT_TRUNCATE = 300;
+const MAX_SUMMARY_LENGTH = 100;
+
+export interface ToolInfo {
+  name: string;
+  input: string;
+  output: string;
+}
+
+export async function generateToolUseSummary(opts: {
+  tools: ToolInfo[];
+  llamaSwapUrl: string;
+  model: string;
+  log: FastifyBaseLogger;
+  signal?: AbortSignal;
+}): Promise<string | null> {
+  const { tools, llamaSwapUrl, model, log, signal } = opts;
+  if (tools.length === 0) return null;
+  if (signal?.aborted) return null;
+
+  const toolText = tools
+    .map(t => `Tool: ${t.name}\nInput: ${t.input.slice(0, INPUT_TRUNCATE)}\nOutput: ${t.output.slice(0, INPUT_TRUNCATE)}`)
+    .join('\n\n');
+
+  try {
+    const res = await fetch(`${llamaSwapUrl}/v1/chat/completions`, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({
+        model,
+        messages: [
+          { role: 'system', content: TOOL_SUMMARY_SYSTEM_PROMPT },
+          { role: 'user', content: toolText },
+        ],
+        max_tokens: 30,
+        temperature: 0.2,
+        stream: false,
+        chat_template_kwargs: { enable_thinking: false },
+      }),
+      signal,
+    });
+    if (!res.ok) {
+      log.debug({ status: res.status }, 'tool-summary: LLM request failed');
+      return null;
+    }
+    const data = await res.json() as { choices?: Array<{ message?: { content?: string } }> };
+    const raw = data.choices?.[0]?.message?.content?.trim() ?? '';
+    if (!raw) return null;
+    // Clean: strip quotes, "Label:" prefix, cap length
+    let cleaned = raw.split('\n')[0]?.trim() ?? '';
+    cleaned = cleaned
+      .replace(/^[-*•]\s+/, '')
+      .replace(/^["'`‘’“”]|["'`‘’“”]$/g, '')
+      .replace(/^(label|summary)\s*:\s*/i, '')
+      .trim();
+    return cleaned.length > MAX_SUMMARY_LENGTH
+      ? cleaned.slice(0, MAX_SUMMARY_LENGTH).trim()
+      : cleaned || null;
+  } catch (err) {
+    log.debug({ err: err instanceof Error ? err.message : String(err) }, 'tool-summary: error');
+    return null;
+  }
+}