3ngram has a native bot for Telegram, Discord, and Slack. WhatsApp is present as a preview surface, but it is not treated as a production-facing channel yet. The bot runs server-side, so you can check in on commitments and context from anywhere without an active coding session.Documentation Index
Fetch the complete documentation index at: https://docs.3ngram.ai/llms.txt
Use this file to discover all available pages before exploring further.
The 3ngram bot (
@EngramAI_bot on Telegram) is distinct from Claude Code Channels, which lets you message an active coding session. The bot works anytime, with shared context for every AI and agent, and does not require a Claude Code session.Setup
- Open Settings > Notifications in the dashboard.
- Use the Bot section to connect Telegram or Discord.
- Start a conversation with
@EngramAI_boton Telegram, or invite the bot to your Discord server. - Send the pairing code shown in the dashboard.
- The bot confirms pairing and is ready to use.
WhatsApp is preview-only. It uses the same registry and budget plumbing as the other platforms, but it is not the public default surface.
Commands
| Command | What it does |
|---|---|
/commitments | List your open commitments |
/blockers | Show current blockers |
/briefing | Start-of-day overview: overdue, stale, recent captures |
/recall <topic> | Semantic search across your memories |
/search <query> | Keyword search (no AI required) |
/ask <question> | AI-synthesized answer from your memories (requires LLM tier) |
/resolve <id> | Mark a commitment as done |
/defer <id> | Defer a commitment to later |
/digest | Your daily digest on demand |
/quiet <start> <end> | Set quiet hours (e.g. /quiet 22:00 08:00) |
/snooze <id> <days> | Snooze a nudge for 1, 3, or 7 days |
/done <id> | Shorthand resolve from a nudge |
/status | Memory and commitment summary |
How it routes
The bot evaluates messages in a fixed order:- Explicit commands, such as
/ask,/resolve, and/budget - Conservative conversational intents
- Generic natural-language questions when AI mode is enabled and budget remains
| Intent | Examples |
|---|---|
| Agenda | ”What’s on my agenda today?” |
| Knowledge synthesis | ”Summarize what I know about recruiting.” |
| Resolve follow-up | ”I already did that” / “finished the onboarding note” |
LLM tier
The/ask command uses an AI model to synthesize answers from your memories and indexed documents. Answers include numbered citations:
Based on [1] and [2]: You chose Railway because of its Postgres support and zero-config deploys.To enable LLM tier:
- Go to Settings > Notifications
- Toggle AI responses on
llm_usage.platform so platform-specific caps can be enforced. If the budget is exceeded, /ask falls back to /search automatically.
Budget enforcement works in layers:
- the user monthly budget gates all bot AI usage first
- the bot policy then applies daily, monthly, soft, hard, and optional per-platform limits
- when a platform cap is hit, AI replies pause on that platform, but commands still work
Proactive nudges
The bot sends proactive nudges without you having to ask:| Nudge type | When it fires |
|---|---|
| Overdue | Every 30 minutes for overdue commitments |
| Due today | Once per day at your local morning |
| Stale | Weekly, for commitments with no activity in 7+ days |
| Digest | Daily at your configured digest time |