Implement task-type router with intelligent model selection and production setup

Major Changes:
- Implement task-type router (src/agents/task-type-router.ts) for intelligent model routing
  * Detects task type from user message (file-analysis, creative, debugging, cli, general)
  * Routes to optimal models: Gemini Flash (file analysis), Llama 3.3 70B (creative),
    Claude Sonnet 4.5 (debugging), Mistral Devstral 2 (CLI/general)
  * Integrated into model selection pipeline for seamless routing

- Integrate task-type routing into model resolution (src/agents/model-selection.ts)
  * Pass userMessage to resolveDefaultModelForAgent for context-aware routing
  * Maintain fallback chain for model availability

- Update attempt runner (src/agents/pi-embedded-runner/run/attempt.ts)
  * Pass prompt context to enable task-type based model selection

- Enhanced security and development (.gitignore)
  * Added comprehensive rules for sensitive files (.env variants, credentials)
  * Excluded API keys, runtime logs, test files, auto-generated skills directories
  * Properly ignored ecosystem.config, build artifacts, package manager locks

- Add technical documentation (README_Tech.md)
  * Process architecture (systemd Gateway, PM2 Dashboard, PM2 AI Product Visualizer)
  * Management commands and troubleshooting guide
  * Configuration summary and deployment checklist
  * Problem log with 6 documented issues and solutions

Result:
- Bot now intelligently routes user requests to optimal models based on message type
- Production-ready with systemd isolation, preventing PM2 conflicts
- Comprehensive documentation for future maintenance and troubleshooting
- Secure version control with quality .gitignore

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
This commit is contained in:
Valtteri Melkko 2026-01-29 15:26:14 +00:00 committed by valtterimelkko
parent 5f4715acfc
commit ab8540870b
5 changed files with 524 additions and 1 deletions

67
.gitignore vendored
View File

@ -71,3 +71,70 @@ USER.md
# local tooling
.serena/
# ===== CUSTOM SECURITY ADDITIONS =====
# Sensitive Configuration & Secrets (NEVER commit these!)
.env
.env.local
.env.*.local
**/.env
**/.env.*
ecosystem.config.cjs
ecosystem.config.js
# API Keys & Credentials
**/auth-profiles.json
**/credentials/
**/secrets/
**/*-credentials.json
**/*-apikey*
**/*-secret*
# System Service Files (local deployment only)
/etc/systemd/system/moltbot-gateway.service
# Runtime & Log Files
/var/log/moltbot-gateway.log
**/moltbot-*.log
**/pm2-*.log
/tmp/moltbot/
/tmp/moltbot-gateway.log
**/logs/
**/log/
*.log
# Test & Validation Files (created during dev)
test-task-router*.js
verify-task-router.js
**/*.test.js
**/*.spec.js
test-results/
test-output/
# Global Skills & Plugins (large, auto-generated)
skills/global-shared/
skills/global-skills/
# OS & IDE Files
.vscode/
.idea/
*.swp
*.swo
*~
.DS_Store
**/.DS_Store
Thumbs.db
# Package Manager Locks (use source control for dependency pinning)
pnpm-lock.yaml
bun.lock
bun.lockb
package-lock.json
# Node/Build
dist/
node_modules/
build/
.cache/
*.bun-build

312
README_Tech.md Normal file
View File

@ -0,0 +1,312 @@
# Moltbot Technical Documentation
## Format Guidelines for Contributors
**Style:** Concise, technical, action-oriented.
**Brevity:** One sentence per command/concept. Use bullet points, not paragraphs.
**Problem Log:** Keep entries short—problem → symptom → solution. Add date and who fixed it if known.
**Commands:** Always include the command first, explanation after (e.g., `systemctl restart moltbot-gateway` # Restarts the gateway service).
**Sections:** Group by topic. Use `##` for major sections, `###` for subsections.
**Updates:** When adding new problems/solutions, add to the end of the Problem Log section with date.
---
## Process Architecture
### Core Components
1. **Moltbot Gateway** (`moltbot-gateway`)
- Service: `/etc/systemd/system/moltbot-gateway.service`
- Runs: `/usr/bin/node dist/entry.js gateway --port 18789`
- Manager: `systemd` (isolated from PM2)
- Handles: Telegram integration, message routing, model selection
2. **Supporting Processes**
- **Dashboard** (si_project/dashboard) - PM2 managed, separate from bot
- **AI Product Visualizer** (ai_product_visualizer) - PM2 managed, separate from bot
- **Telegram Relay** - Embedded in gateway (grammY framework)
- **Task-Type Router** - Compiled TypeScript module in gateway
3. **Configuration Files**
- Global: `/root/.clawdbot/moltbot.json`
- Agent-specific: `/root/.clawdbot/agents/main/config.json`
- Environment: `/root/.clawdbot/.env`
---
## Process Management
### Moltbot Gateway (Systemd)
```bash
# Check status
systemctl status moltbot-gateway
# Restart (reloads config + code)
systemctl restart moltbot-gateway
# Stop gracefully
systemctl stop moltbot-gateway
# Start if stopped
systemctl start moltbot-gateway
# View live logs
journalctl -u moltbot-gateway -f
# View last 100 lines
journalctl -u moltbot-gateway -n 100
```
**Auto-restart:** Enabled. If process crashes, systemd restarts it within 5 seconds.
**Boot persistence:** Enabled. Starts automatically on system reboot.
### From Telegram Chat
Send `/restart` command in Telegram to restart the bot gracefully without terminal access.
### Dashboard (PM2)
```bash
# Check status
pm2 list
# Restart
pm2 restart dashboard
# Logs
pm2 logs dashboard
# Stop
pm2 stop dashboard
```
**Isolation:** Runs in separate PM2 daemon. Does not interfere with Moltbot.
### Logs Location
```bash
# Moltbot systemd logs
journalctl -u moltbot-gateway -n 200
# Moltbot app logs (most detailed)
tail -f /var/log/moltbot-gateway.log
# Application debug logs
tail -f /tmp/moltbot/moltbot-*.log
```
---
## Problem Log & Solutions
### 1. **Duplicate Telegram Responses** (Jan 28, 2026)
**Problem:** Bot sending same message 2-3 times.
**Root Cause:** `streamMode: "partial"` in Telegram config caused responses to stream as chunks, each sent separately.
**Solution:** Changed `streamMode` from `"partial"` to `"block"` in `/root/.clawdbot/moltbot.json`.
```json
"telegram": {
"streamMode": "block" // Single unified message
}
```
**Status:** ✅ Fixed. Single responses now.
---
### 2. **Unknown Model Error** (Jan 28, 2026)
**Problem:** Error: `Unknown model: openrouter/mistralai/mistral-devstral-2`
**Root Cause:** Incorrect OpenRouter model ID format. Used old naming convention.
**Solution:** Updated model IDs to correct OpenRouter format:
- `mistralai/devstral-2512` (Mistral Devstral 2)
- `google/gemini-2.0-flash-001` (Gemini 2.0 Flash)
- `meta-llama/llama-3.3-70b-instruct:free` (Llama 3.3 70B)
**Status:** ✅ Fixed. Models now load correctly.
---
### 3. **PM2 Process Isolation Conflict** (Jan 28, 2026)
**Problem:** Dashboard PM2 restarting 140+ times. Gateway conflicting with dashboard in same PM2 daemon.
**Root Cause:** Moltbot gateway was added to default PM2 instance, sharing resources with dashboard.
**Solution:** Moved Moltbot from PM2 to systemd service (isolated).
- Moltbot: `systemd` only
- Dashboard: `PM2` only
- No shared daemon = no conflicts
**Status:** ✅ Fixed. Processes now isolated.
**Files changed:**
- Created: `/etc/systemd/system/moltbot-gateway.service`
- Removed: Moltbot from PM2 list
---
### 4. **Missing Task-Type Router Compilation** (Jan 28, 2026)
**Problem:** Bot said it implemented task-type routing but nothing changed.
**Root Cause:** TypeScript source files modified but not compiled to `dist/`.
**Solution:**
1. Fixed import error in `src/agents/task-type-router.ts` (DEFAULT_PROVIDER location)
2. Compiled: `npm run build`
3. Restarted gateway to load new `dist/` code
**Status:** ✅ Fixed. Task-type router now active.
---
### 5. **Telegram Command Limit Exceeded** (Jan 29, 2026)
**Problem:** Error: `setMyCommands failed: BOT_COMMANDS_TOO_MUCH` (Telegram API limit = 100 commands).
**Root Cause:** Both config files had `"native": "auto"` trying to register all skills + commands with Telegram.
**Solution:** Disabled native command auto-registration:
```json
// /root/.clawdbot/moltbot.json
"commands": {
"native": false,
"nativeSkills": false
}
// /root/.clawdbot/agents/main/config.json
"commands": {
"native": false,
"text": true,
"restart": true
}
```
**Status:** ✅ Fixed. Telegram now connects without errors.
---
### 6. **Node.js Version Too Old** (Jan 28, 2026)
**Problem:** Moltbot requires Node.js 24+ but only v20 was installed.
**Root Cause:** Package.json specified `engines: { node: ">=24" }`.
**Solution:** Upgraded Node.js:
```bash
curl -fsSL https://deb.nodesource.com/setup_24.x | sudo -E bash -
sudo apt-get install -y nodejs
```
**Verified:** `node --version` → v24.13.0
**Status:** ✅ Fixed.
---
## Configuration Summary
### Model Fallback Chain
**Primary:** Mistral Devstral 2 2512 (agentic specialist)
**Fallbacks:**
1. Gemini 2.0 Flash (long-context, 1M tokens)
2. Llama 3.3 70B (creative/pedagogical)
3. Moonshot Kimi K2.5 (language model)
4. Claude Sonnet 4.5 (escalation)
5. Claude Opus 4.5 (complex reasoning)
### Task-Type Routing
- **File Analysis** → Gemini Flash
- **Creative Content** → Llama 3.3 70B
- **Debugging** → Claude Sonnet 4.5
- **CLI/Commands** → Mistral Devstral 2
- **General** → Mistral Devstral 2 (default)
### Telegram Settings
- **Streaming Mode:** `block` (single message per response)
- **Commands Native:** `false` (avoid API limit)
- **Restart Command:** `true` (allows `/restart` from chat)
- **User ID Allowlist:** 876311493 (only you)
---
## Quick Troubleshooting
### Bot Not Responding
1. Check status: `systemctl status moltbot-gateway`
2. Check logs: `journalctl -u moltbot-gateway -n 50`
3. Restart: `systemctl restart moltbot-gateway`
4. Verify Telegram: `node dist/entry.js channels status`
### Telegram Connection Error
Check logs for `setMyCommands failed` or network errors.
If command limit error: Verify `native: false` in both config files.
### High Latency (>1 minute)
Expected for first API call to OpenRouter. Check OpenRouter API status.
If consistent, check model health: `node dist/entry.js models status`
### Duplicate Responses
Check `streamMode: "block"` is set in `/root/.clawdbot/moltbot.json`.
If issue persists, reduce retry attempts in retry policy config.
---
## Deployment Checklist
- [ ] Node.js 24+ installed
- [ ] Moltbot cloned and built (`npm run build`)
- [ ] Systemd service created and enabled
- [ ] Config files populated (moltbot.json, agents/main/config.json)
- [ ] API keys in environment or .env
- [ ] Telegram bot token configured
- [ ] Gateway started: `systemctl start moltbot-gateway`
- [ ] Telegram connection verified: `node dist/entry.js channels status`
- [ ] Test message sent in Telegram
---
## Key File Locations
```
/root/moltbot/ Main installation
├── dist/ Compiled code (loaded at runtime)
├── src/ TypeScript source
├── ecosystem.config.cjs PM2 config (legacy, not used)
└── README_Tech.md This file
~/.clawdbot/ Config directory
├── moltbot.json Global gateway config
├── agents/main/
│ ├── config.json Agent-specific config
│ └── auth-profiles.json API key storage
└── .env Environment variables
/etc/systemd/system/ System services
└── moltbot-gateway.service Systemd service file
/var/log/ System logs
└── moltbot-gateway.log Gateway application log
/tmp/moltbot/ Runtime logs
└── moltbot-*.log Detailed debug logs
```
---
**Last Updated:** Jan 29, 2026
**Maintained By:** Claude Code + Moltbot Task Router

View File

@ -3,6 +3,7 @@ import type { ModelCatalogEntry } from "./model-catalog.js";
import { normalizeGoogleModelId } from "./models-config.providers.js";
import { resolveAgentModelPrimary } from "./agent-scope.js";
import { DEFAULT_MODEL, DEFAULT_PROVIDER } from "./defaults.js";
import { resolveModelForAgentWithTaskRouting, type TaskType } from "./task-type-router.js";
export type ModelRef = {
provider: string;
@ -156,6 +157,7 @@ export function resolveConfiguredModelRef(params: {
export function resolveDefaultModelForAgent(params: {
cfg: MoltbotConfig;
agentId?: string;
userMessage?: string;
}): ModelRef {
const agentModelOverride = params.agentId
? resolveAgentModelPrimary(params.cfg, params.agentId)
@ -178,11 +180,21 @@ export function resolveDefaultModelForAgent(params: {
},
}
: params.cfg;
return resolveConfiguredModelRef({
// Get the default model using original logic
const defaultModelRef = resolveConfiguredModelRef({
cfg,
defaultProvider: DEFAULT_PROVIDER,
defaultModel: DEFAULT_MODEL,
});
// Apply task-type routing if user message is provided
return resolveModelForAgentWithTaskRouting({
cfg: params.cfg,
agentId: params.agentId,
userMessage: params.userMessage,
defaultModelRef,
});
}
export function buildAllowedModelSet(params: {

View File

@ -305,6 +305,7 @@ export async function runEmbeddedAttempt(
const defaultModelRef = resolveDefaultModelForAgent({
cfg: params.config ?? {},
agentId: sessionAgentId,
userMessage: params.prompt,
});
const defaultModelLabel = `${defaultModelRef.provider}/${defaultModelRef.model}`;
const { runtimeInfo, userTimezone, userTime, userTimeFormat } = buildSystemPromptParams({

View File

@ -0,0 +1,131 @@
import type { MoltbotConfig } from "../config/config.js";
import type { ModelRef } from "./model-selection.js";
import { parseModelRef } from "./model-selection.js";
import { DEFAULT_PROVIDER } from "./defaults.js";
export type TaskType = "file-analysis" | "creative" | "debugging" | "cli" | "general";
/**
* Detect task type from user message using keyword analysis
*/
export function detectTaskType(userMessage: string): TaskType {
if (!userMessage || typeof userMessage !== "string") {
return "general";
}
const messageLower = userMessage.toLowerCase();
// File analysis tasks - route to Gemini for large context window
if (
messageLower.includes("file") ||
messageLower.includes("read") ||
messageLower.includes("analyze") ||
messageLower.includes("directory") ||
messageLower.includes("folder") ||
messageLower.includes("content of") ||
messageLower.includes("examine") ||
messageLower.includes("inspect") ||
messageLower.includes("scan") ||
messageLower.includes("browse") ||
messageLower.includes("list files") ||
messageLower.includes("show files")
) {
return "file-analysis";
}
// Creative content tasks - route to Llama for stylistic versatility
if (
messageLower.includes("write") ||
messageLower.includes("create") ||
messageLower.includes("content") ||
messageLower.includes("story") ||
messageLower.includes("poem") ||
messageLower.includes("article") ||
messageLower.includes("compose") ||
messageLower.includes("draft") ||
messageLower.includes("generate") ||
messageLower.includes("summarize") ||
messageLower.includes("explain") ||
messageLower.includes("describe")
) {
return "creative";
}
// Debugging tasks - route to Claude for complex reasoning
if (
messageLower.includes("debug") ||
messageLower.includes("error") ||
messageLower.includes("fix") ||
messageLower.includes("troubleshoot") ||
messageLower.includes("complex") ||
messageLower.includes("problem") ||
messageLower.includes("issue") ||
messageLower.includes("bug") ||
messageLower.includes("broken") ||
messageLower.includes("not working") ||
messageLower.includes("failed") ||
messageLower.includes("crash")
) {
return "debugging";
}
// CLI/terminal tasks - route to Mistral for agentic workflows
if (
messageLower.includes("terminal") ||
messageLower.includes("command") ||
messageLower.includes("cli") ||
messageLower.includes("exec") ||
messageLower.includes("bash") ||
messageLower.includes("shell") ||
messageLower.includes("run") ||
messageLower.includes("execute") ||
messageLower.includes("script") ||
messageLower.includes("install") ||
messageLower.includes("update") ||
messageLower.includes("upgrade")
) {
return "cli";
}
return "general";
}
/**
* Map task types to optimal models based on your LLM strategy
*/
export function resolveModelForTaskType(taskType: TaskType, cfg: MoltbotConfig): ModelRef | null {
// Define the optimal model mapping based on your strategy
const TASK_TYPE_MODEL_MAPPING: Record<TaskType, string> = {
"file-analysis": "openrouter/google/gemini-2.0-flash", // 1M context for spatial reasoning
creative: "openrouter/meta-llama/llama-3.3-70b-instruct", // Best for pedagogical content
debugging: "anthropic/claude-sonnet-4-5", // Complex reasoning specialist
cli: "openrouter/mistralai/mistral-devstral-2", // Agentic workflow specialist
general: "openrouter/mistralai/mistral-devstral-2", // Default to agentic specialist
};
const modelRef = TASK_TYPE_MODEL_MAPPING[taskType];
return parseModelRef(modelRef, DEFAULT_PROVIDER);
}
/**
* Enhanced model resolution that considers task type for optimal routing
*/
export function resolveModelForAgentWithTaskRouting(params: {
cfg: MoltbotConfig;
agentId?: string;
userMessage?: string;
defaultModelRef: ModelRef;
}): ModelRef {
// If we have a user message, use task-type routing
if (params.userMessage) {
const taskType = detectTaskType(params.userMessage);
const taskModel = resolveModelForTaskType(taskType, params.cfg);
if (taskModel) {
return taskModel;
}
}
// Fall back to the default model
return params.defaultModelRef;
}