Introducing /agent - on-demand specialist AI in your server
A new Jarvis slash command that picks the right specialist AI for whatever you throw at it. Code, research, math, creative work - one command, the right model.
/agent is live in every server that has Jarvis. Type a short brief, hit enter, and the right specialist AI runs the task and reports back in Jarvis's voice. No model picking, no dropdown menus, no setup.
What it does
/agent <brief> takes up to 200 characters of plain English describing what you want. Jarvis picks the model best suited for the job, runs the task, and replies in his usual tone with the result.
A few examples that fit in 200 characters:
- Write a TypeScript function that debounces a callback by 250ms.
- Compare HTTP/2 and HTTP/3 in two bullets.
- Draft a one-line noir intro for a butler character.
- Find the latest stable React release and summarise the changes.
Different briefs route to different specialists - code, research, creative, math - automatically. You don't see the routing; you just get the answer.
Why a separate command
/jarvis is for conversation: low latency, chatty. /agent is for tasks: one shot, slower, pickier about which model runs it. If you want to chat, use /jarvis. If you want a task done well, use /agent.
Code comes back as a file
Long code outputs are attached as a downloadable file with the right extension - .py, .ts, .go, whatever the agent emitted. Short replies stay inline. The cutoff is roughly six lines of code or 800 characters.
What it isn't
- One-shot only. No follow-ups. Re-run with a new brief.
- Not a code editor. It writes code at you; it doesn't touch your repo.
- Server only. Doesn't work in DMs or group DMs.
- Not callable from
/jarvischat. Only the slash command triggers it.
The limits
- 200-character brief.
- 60-second cooldown per user, per server.
- 30 calls per user per day, 30 per server per day.
- One in-flight per user.
- 90-second hard timeout on the agent itself.
Safety
Briefs are screened for jailbreaks and obvious exfil. The agent runs sandboxed - it can't see Jarvis's tokens, server data, or your messages. Output is scrubbed before it reaches Discord. Calls are logged for 30 days for moderation; the log is local and not exposed.
If you find a way through any of this, email dev@jorvis.org.
Try it
In any server with Jarvis, type /agent and describe one thing you want done. Keep it short - clearer briefs route to better specialists. Bug reports go to the AGIS support server.
