Jambot 8.1.4, from install to Deep Work.

Quick setup

1. Install Jambot with Administrator or permission integer 10292755164407.
2. Move Jambot's role above the roles and members it needs to manage.
3. Run /permissions-check and add staff roles with /access add-role.
4. Configure verification, raid protection, tickets, or other server tools as needed.
5. GrandJam stays off until a Jambot admin grants it.

GrandJam AI

GrandJam is disabled by default for every guild. A Jambot admin grants access with /jambot-admin grant-grandjam. After it is granted, authorized server staff use /grandjam console to control AI chat, passive replies, Tickets Only mode, allowed channels, Deep Work behavior, model preference, pass count, and file output.

/grandjam prompt can upload a UTF-8 .txt, .md, or .markdown server prompt up to 50 KiB. These server instructions are included in new AI turns immediately but remain below core safety and privacy rules.

Deep Work and extended thinking

Deep Work can be Auto, Always, or Off. Auto keeps simple questions fast and opens a work thread for larger or concrete build requests. Coding and project-building requests are treated as artifact work when file delivery is enabled.

The status changes to match the operation currently running. Stages that are not needed are skipped. A research task can begin with Researching, while a local build can start at Planning.

The planning pass must restate what the user asked for and produce only a high-level execution plan. It is not allowed to output final source code, file contents, commands, or completion claims. If a model starts solving the task during planning, Jambot discards that plan and retries with a stricter plan-only instruction before falling back to a safe local plan summary.

With three passes, Jambot plans, creates an internal first build, then reviews or rebuilds it before the final response. The thread shows concise summaries and progress, never hidden chain-of-thought.

Code and file delivery

When a request calls for a real project or code file, Jambot can package the result as Discord attachments instead of splitting the implementation across multiple chat messages. A Flask calculator can finish with an app.py attachment; multi-file projects can send several files or a ZIP when needed.

If the model returns a good solution but fails to package the artifacts, Jambot can run an extra repackaging pass that converts the finished output into files. Generated attachments can also be mirrored into the private AI audit record so moderators can see the same files the user received.

Web research

GrandJam can research explicit requests such as "research this", "search the web", "look this up", and "find sources". It can also search informational or freshness-sensitive questions when automatic research is enabled. Search uses DuckDuckGo result pages and can fall back between supported result layouts.

When research is part of Deep Work, the thread begins with Researching and moves to Planning after current context is gathered. If search fails or returns no usable context, Jambot is instructed not to pretend that research succeeded.

Cybersecurity help and suspicious-request audit

GrandJam is meant to answer harmless educational and defensive cybersecurity questions, including what malware, RATs, keyloggers, phishing, and similar concepts are, how to identify risks, and how to protect or clean a system.

Clearly action-oriented abusive requests can be refused and privately copied to the suspicious-users record channel for human review. That AI audit does not automatically punish the member. Suspicious requests also do not trigger automatic web research.

Jambot also detects obviously degenerate model output such as enormous repeated punctuation runs. It can retry with the other provider and suppress the broken output instead of flooding Discord.

Global AI and search acknowledgement

Before Jambot sends a member's content to AI or performs DuckDuckGo search, the member must acknowledge the privacy notice. The choice is global for that Discord user across servers and DMs. /privacy shows the current status and allows the user to change it.

AI prompts, replies, search records, errors, safety events, and generated attachments can be copied into Jambot's configured private record server. Normal users receive up to 5 AI DMs per UTC day; Jambot admins and specifically rewarded users bypass that limit.

Support tickets

/tickets setup chooses support roles plus optional category, panel channel, and ticket log channel. The ticket panel creates one private ticket per member, gives access to the opener and selected support roles, and can ping those roles when the ticket opens.

Ticket closing uses Continue and Cancel before the ticket channel is deleted. When GrandJam is configured for Tickets Only, AI is available only inside Jambot ticket channels.

Public action confirmations

Sensitive moderator actions can post a clean Continue and Cancel confirmation that everyone may see while only the authorized moderator can interact with it. This covers moderation and destructive workflows such as bans, kicks, warnings, timeouts, purges, role changes, ticket closing, clean scans, and scoped test operations.

Say and embed builder are separate

/say sends plain text as Jambot. /embed-builder opens the interactive embed builder separately. The prefix version ?say follows the same plain-message idea.

Jambot administrator controls

These commands are restricted to IDs configured as Jambot admins. User-facing bot text does not expose the internal configuration label.

• /jambot-admin servers opens a paginated guild list with names, guild IDs, and member counts.
• The server browser and /jambot-admin invite can create a one-use invite that expires after 60 seconds.
• /jambot-admin grant-grandjam and /jambot-admin revoke-grandjam control server access to GrandJam.
• /jambot-admin reward-dm and /jambot-admin remove-dm-reward manage AI DM rewards.
• /jambot-admin nuke-test creates tagged test channels only after confirmation, and cleanup is scoped to those tagged channels.

The required-permissions install link includes Create Instant Invite so the temporary invite browser works without Administrator.

AI providers and diagnostics

Jambot supports Cloudflare Workers AI and Groq with multiple configured credentials, provider failover, optional extended models for Deep Work, and a cooldown when an extended model is unavailable. Normal chat can keep the fast model configuration while Deep Work prefers the separately configured stronger model when enabled.

/grandjam diagnostics can test the normal or deep models without displaying credentials. AI failures receive a short Error ID, and /grandjam error retrieves the saved redacted failure detail.

Verification

1. A member presses the verification button in Discord.
2. Jambot creates a one-time server-bound request for that member.
3. The website exchanges the one-time key for a signed HTTP-only session and validates Cloudflare Turnstile.
4. The website marks a successful security check ready. It does not assign roles.
5. The bot rechecks membership, current server settings, permissions, and role hierarchy before changing roles.

No Discord OAuth sign-in is requested for verification. Use /verify setup and run /verify test before relying on it.

Moderation, raid protection, and backups

Warnings, persistent mutes, user history, server settings, verification configuration, and backups remain guild-scoped. Raid protection can react to message bursts, restore deleted channels when possible, and remove spam-created channels. Backup and restore operations remain limited to the current guild boundary.