An AI chat bot for Discord with memory and personality
Cortex turns your bot into a member of the community. Mention it and it answers like a person — server questions, banter, explanations, recommendations — with memory of the conversation and the server, so it doesn't reintroduce itself every message. Give it a persona: formal concierge, chaotic gremlin, or in-character for your game — you write the personality, Cortex plays it.
Start freeStart free. Pro is $9.99/month.
It runs on your bot, under your name and avatar, in every server your bot is in. You can also turn on an optional channel-activity rule: Cortex may choose a text reply to a regular message in the channels you select, from members with the roles you select. It stays off until you switch it on and save. You choose the wait between replies, and that wait continues across bot restarts; a silent draw can leave the channel quiet. Keyword rules, @Cortex mentions, replies to Cortex, messages in the Cortex channel and game answers keep their normal handling. Cortex AI is included in Pro ($9.99/mo — 3 bots, all 49 modules, plus the AI Module Builder).
What you get
Opt-in message forwarding
Copy new member text between selected text channels in the same server, with an original-message link and a wait between attempts. Off by default; no AI call, attachments or mention notifications.
Wildcard reply patterns
Opt in to full-message, case-insensitive patterns for reusable text replies. help * matches help music; *help* matches help anywhere. Uses the same reply variants and cooldowns, with no AI call.
Chats like a member
Mention your bot and Cortex answers naturally — questions, banter, server help, anything.
What it needs
- Enable Cortex AI on your bot. Custom replies need View Channel, Send Messages and Message Content Intent; these plain-text replies do not make an AI call.
- Configure replies for the selected server in the dashboard. Auto-responders created with !cortexautorespond remain saved with their existing behavior.
Setting it up
- Open your bot’s Modules tab, configure Cortex AI and select a server. Keep Enable auto-responses on and add a trigger. Choose Contains, Exact message or Starts with for case-insensitive keywords, or Reply to channel activity for ordinary messages. This option is off until a manager selects it, chooses 1–25 text channels and 1–25 member roles, and saves. A member needs at least one selected role. It never takes over keyword matches, @Cortex mentions, replies to Cortex, messages in the Cortex channel or game answers. Threads are excluded.
- For a reusable pattern rule, choose Wildcard pattern and enter help *. It matches the whole message, ignoring case: * means zero or more characters and ? means one Unicode code point, including a line break. Escape literal symbols as \*, \? and \\. All other characters are literal, not regex syntax. Add a reply such as See the help channels, then Save Settings. Existing rules keep their match type; generated bots need regeneration to receive this matcher.
- Add up to 10 reply variants. A blank weight means 1; otherwise use a positive whole number. For an activity rule, a positive No reply weight makes some draws stay silent. Set its required Cooldown (seconds) from 1 to 86,400 and choose Per channel or Across selected channels. A sent reply starts the wait, which continues across bot restarts; a silent draw does not start or extend it. Save Settings.
- For message forwarding, open Message forwarding in the same server’s Cortex settings. Add a route, choose different source and destination text channels, set a 1–86,400 second wait, enable the route and Save Settings. Check who can read the destination before enabling a private source. This runs independently of auto-responses. Existing generated bots need regeneration.
Example
To let Cortex sometimes reply in #general and #lounge, choose Reply to channel activity, select those channels and a member role, add “Glad you’re here!” with weight 1, add No reply weight 49, set 600 seconds and choose Across selected channels. Each eligible regular message gets a chance; most draws stay quiet. After Cortex sends the reply, all selected channels wait ten minutes, and that wait survives a bot restart. With Per channel, only the channel where Cortex replied waits. Keyword matches, @Cortex mentions, replies to Cortex, messages in the Cortex channel and game answers are handled normally and do not use this activity rule.
Limits and troubleshooting
Limits
- Use up to 50 dashboard triggers per server, with 1–10 plain-text replies per trigger and up to 2,000 characters per reply. Missing reply weights are equal; a missing or zero no-reply weight is disabled; missing cooldown or 0 means no cooldown. {user}, {server} and {channel} placeholders are supported, without mention notifications.
- Wildcard patterns allow 1–200 characters, require a non-space literal character and ignore messages longer than 4,000 UTF-16 code units. Catch-all patterns such as * are rejected; use the separate activity option for ambient replies. There are no capture substitutions, loops or AutoMod actions in this pattern matcher. Auto-responses never forward; use Message forwarding for that. Weighted replies, silent outcomes, first-match order and per-channel/per-user cooldowns work as for keyword rules.
- The first matching keyword rule runs before activity replies, legacy auto-responders and AI chat. A deliberate silent outcome also stops there. Keyword cooldowns reset on restart or rule change by default. Activity waits begin when a reply is chosen, before sending to Discord, and remain saved through a restart; they stay reserved even if Discord rejects the send. If saved cooldown storage is unavailable, activity replies stay silent.
- Activity replies ignore bots, webhooks, system messages, empty messages and known bot commands. This feature does not collect message history or call an external provider. Its cooldown table stores only server/channel IDs (or a server scope marker), an opaque rule hash and expiry, never message text or member IDs. Expired and removed-rule entries are pruned on the next eligible message in that server; rule edits reset the wait. Other Cortex features retain their own settings.
- Forwarding allows up to 10 routes per server, one destination per source. It copies at most 4,000 text characters and a link to the original message. Messages during the wait are skipped; waits start before each attempt and reset on restart or route edits. Bots, webhooks, system messages, known bot commands, threads and announcement channels are excluded. No attachments, history, edit/delete sync or retries. Disabling a route stops future copies; existing copies remain. The bot needs View Channel at the source and View Channel, Send Messages and Embed Links at the destination.
If it doesn’t work
- No reply: existing generated bots need regeneration to receive activity replies. Check that Cortex and Enable auto-responses are enabled, the correct server is selected, the message is in an allowed channel, the member has an allowed role, and the bot has channel access and Message Content Intent.
- A regular message stays quiet: the draw may have chosen the silent outcome, or the selected channel/server may still be waiting. Check both lists and the required 1–86,400 second wait. Activity is intentionally off until you select it and save.
Example commands
| Command | What it does |
|---|---|
!cortex | View Cortex AI settings. |
!cortexpersonality Friendly and concise | Set the personality used for Cortex responses. |
!cortexreset | Clear conversation history for the current channel. |
Full command list and configuration options in the Cortex AI documentation.
What does it cost?
Cortex AI is included in Pro — Start free. Pro is $9.99/month.
- 1 hosted bot, online 24/7
- 20 free-forever modules
- 10 custom commands
- No credit card required
- 3 bots
- All 49 modules — including Cortex AI
- AI Module Builder (50 credits/mo)
Frequently asked questions
Is there a free AI chat bot for Discord?
Cortex AI is a Pro module ($9.99/mo — 3 bots, all 49 modules, AI Module Builder included).
Do I need API keys or code?
Neither. The AI runs through the platform — no OpenAI keys to manage, no code to write, no servers to rent. Persona and memory settings are plain-language configuration, and conversations stay scoped to your bot's own database.
Can Cortex forward messages to another channel?
Yes, through explicit Message forwarding routes in the dashboard. Select a server, choose one source and one different destination text channel, enable the route and save. Copies include new member text (up to 4,000 characters) and an original-message link, without mention notifications. Check destination access: readers there can read copied private-channel text. Up to 10 routes, one destination per source, with a 1–86,400 second wait after every attempt; messages during that wait are skipped. Waits reset on restart or rule edits. No bots, webhooks, system messages, known bot commands, threads, announcements, attachments, history, edit/delete sync or retries. Auto-responses can stay off. Requires source View Channel and destination View Channel, Send Messages and Embed Links. Off by default; existing bots need regeneration.
Related modules
Add Cortex AI to your server today
Create your bot, toggle the module on, and it's live — hosted for you, no code, no downloads.
Start freeStart free. Pro is $9.99/month.
