Skip to main content
The AI Response feature registers a /ai slash command that generates a GPT-powered reply directly inside a Discord thread. The bot reads the thread’s message history, optionally accepts a custom prompt, then streams a response using the gpt-5.2-codex model. Only members with the Manage Messages permission can invoke the command.
The /ai command only works inside threads (announcement, public, or private). Using it in a regular channel returns an ephemeral error message.

​
How it works

1

Run /ai in a thread

Use the /ai command, optionally passing a prompt string and a reasoning level (low, medium, or high). The default reasoning level is medium.
2

Bot starts thinking

The bot replies with an ephemeral message — “The clanker is thinking…” — and a Cancel button. The message updates with a tool-call counter as the model works.
3

Review the response

When the model finishes, the ephemeral message is replaced with the generated text and an Accept button.
4

Post or discard

Click Accept to post the response as a regular message in the thread, or Cancel at any time to stop generation and dismiss the ephemeral message.

​
AI tools

The model has access to three tools that let it read and search the effect-smol repository at generation time.

read

Reads a file (or a line range) from the Effect repository.

rg

Runs ripgrep inside the repository to search for patterns.

glob

Finds files matching a glob pattern inside the repository.
AiResponse.ts
const Tools = Toolkit.make(
  Tool.make("read", {
    description: "Read a file from the effect repository",
    parameters: Schema.Struct({
      path: Schema.String.annotate({
        description:
          "The path to the file to read, relative to the root of the repository",
      }),
      startLine: Schema.optionalKey(Schema.Number).annotate({
        description: "The line number to start reading from (inclusive)",
      }),
      endLine: Schema.optionalKey(Schema.Number).annotate({
        description: "The line number to stop reading at (exclusive)",
      }),
    }),
    success: Schema.String,
  }),
  Tool.make("rg", {
    description: "Wrapper around the 'rg' command.",
    parameters: Schema.Struct({
      pattern: Schema.String.annotate({
        description: "The pattern to pass to rg.",
      }),
      glob: Schema.optionalKey(Schema.String).annotate({
        description: "The --glob option to rg",
      }),
      maxLines: Schema.Finite.annotate({
        description:
          "The total maximum number of lines to return across all files",
      }),
    }),
    success: Schema.String,
  }),
  Tool.make("glob", {
    description: "Find files in the effect repository matching a glob pattern",
    parameters: Schema.Struct({
      pattern: Schema.String.annotate({
        description: "The glob pattern to match files against (e.g. '**/*.ts')",
      }),
    }),
    success: Schema.String,
  }),
)

​
Effect repository context

The bot clones effect-smol on startup and pulls the latest changes every 15 minutes so responses always reflect current source code.
EffectRepo.ts
// Pull the repo every 15 minutes to keep it up to date
yield* Effect.gen(function* () {
  const repoPath = yield* repo
  while (true) {
    yield* Effect.sleep("15 minutes")
    yield* git.pull(repoPath)
    yield* RcRef.invalidate(llmsMd)
  }
}).pipe(Effect.retry(Schedule.forever), Effect.forkScoped)
The system prompt also includes the full contents of LLMS.md from the repository root so the model understands the project’s API conventions before diving into individual files.

​
Models

PurposeModel
Response generationgpt-5.2-codex
Title generation (AutoThreads)gpt-5.2

​
Configuration

Environment variableRequiredDescription
OPENAI_API_KEYYesOpenAI API key used for all AI features.

​
Permissions

The /ai command requires the Manage Messages (ManageMessages) permission. Members without this permission cannot invoke it.

Build docs developers (and LLMs) love

Get started for freeTalk to us