Skip to main content

​
User Guide

Gigi is in a foundation rebuild. The current runtime can log in to Discord when enabled, sync /ping, answer DM or guild-mention ping messages, let authorized admins manage role/user capability grants through /permissions, manage guild-scoped LLM credentials, model profiles, and routing policy through /llm, route /ask through the agent runtime with context:none or dynamic current-channel context fetching through context:channel, opportunistically fetch current-channel context for guild mentions when memory permission allows it, manage durable runs through /agent, show aggregate guild agent stats through /agent stats guild, enable or disable the optional reply latency footer through /agent stats guild reply-latency:on|off, dry-run match enabled external app prefix triggers, route flexible guild mentions through a configured routing model when guild policy allows it, answer unmatched guild mentions through a configured chat model, dispatch consented public send_message prefix triggers declared by approved manifests, build cited context-pack metadata for permitted current-channel memory, persist durable agent run/step/confirmation records, and use current-channel guild memory count/search for enabled channels. Rich DM chat, reasoning chat, restricted external app dispatch, tasks, and relay actions are not live yet.

​
Current Live Checks

  • GET /healthz: process is alive and reports build metadata.
  • GET /readyz: process is configured and can reach PostgreSQL.
  • /ping: replies pong in Discord when command sync is enabled.
  • /permissions: creates/assigns Discord roles and grants or revokes role/user capabilities for authorized admins.
  • /llm: lists, adds, tests, rotates, deletes, and selects guild-scoped provider credentials and model profiles when GIGI_LLM_SECRET_KEY_BASE64 is configured for credential entry.
  • /ask: sends a question through the agent runtime. context:none avoids memory fetches; context:channel can fetch permitted retained messages from the current channel and answer from bounded context/tool results.
  • /agent trace last: shows your last agent planner/tool/answer path in the current channel using safe metadata only.
  • /agent stats guild: shows aggregate server-level agent run, latency, step, tool-call, LLM-call, status, termination, and top-tool summaries for 24h, 7d, 30d, or all when the requester has agent.analytics; it does not show raw prompts, snippets, provider payloads, tool arguments, user IDs, or per-user details.
  • /agent stats guild reply-latency:on|off: enables or disables the optional reply latency footer for future guild replies when the requester has agent.reply_latency.manage. The footer shows only response elapsed wall time. It does not show token counts, provider usage, raw prompts, snippets, or per-user details. Use /llm usage guild for aggregate token/provider usage.
  • /agent cancel, /agent pending, /agent confirm, /agent reject: manage durable runs and pending confirmation records for the run owner or a Discord administrator.
  • Guild mention external app matching: replies with a dry-run plan, or sends the planned command when an enabled public manifest declares dispatch: "send_message" and has stored public-dispatch consent.
  • Guild mention semantic routing: after deterministic matching fails, a configured routing model can propose validated native memory or manifest-grounded plans when /llm routing is dry-run or enabled.
  • Guild mention context: Gigi attempts current-channel context when available, but missing memory permission should not block normal non-memory chat.
  • Guild mention chat fallback: after plugin routing fails, a configured chat model can answer the mention.
  • DM ping: replies pong.
  • Other DM text: replies that rich chat needs a server LLM profile first; rich DM chat is not live.
  • Guild mention @Gigi ping: replies pong.

​
Agent Smoke Test

Use this flow after configuring a guild provider, chat model, routing model, memory permission, and /llm routing set mode:enabled:
The trace should show whether Gigi planned tools such as memory.recent, executed them, and used the answerer. Guild stats should show aggregate run health only if you have agent.analytics. Use /agent stats guild reply-latency:on to opt in to the elapsed-time footer, and /agent stats guild reply-latency:off to disable it. Switch /llm routing set mode:dry-run to check planning without executing tools.

​
Planned Discord Behavior

Future Discord slices are expected to restore and expand:
  • rich DM conversation
  • richer reasoning chat
  • more structured commands
  • restricted or confirmed external app command dispatch declared by approved installed manifests
  • admin-installed external app capabilities
  • permission-aware task, relay, memory, and retrieval features
These richer behaviors are not current shipped behavior in this foundation slice.

​
Next Pages