reference13 tools

Using the agent

SILV runs on Claude Opus 4.8 with adaptive thinking and an agentic tool loop — you describe an outcome in plain English, and it chains the tools below to get there. You never call tools yourself: say “reply to anything mentioning support” and SILV sets up the monitor for you.

Tool reference

Everything the agent can do. Platform tools appear only when that platform is connected.

toolplatformwhat it does
discord_list_channelsDiscordLists the servers and channels the bot can see.
discord_read_messagesDiscordFetches up to 100 recent messages from a channel.
discord_send_messageDiscordSends a message to a channel.
discord_replyDiscordPosts a threaded reply to a specific message.
telegram_read_messagesTelegramReads from the live in-memory buffer of messages received since the bot connected (last 500 kept — bots can’t fetch older history).
telegram_send_messageTelegramSends a message to a numeric chat id, or @channelusername for channels the bot administers.
telegram_replyTelegramReplies to a specific Telegram message.
create_monitorBothWatches incoming messages for a match pattern and runs your instructions automatically when one hits.
list_monitorsBothLists the currently active monitors.
delete_monitorBothRemoves a monitor.
create_scheduled_taskBothRuns instructions on a schedule — daily at HH:MM (server-local) or every N minutes.
list_scheduled_tasksBothLists the currently active scheduled tasks.
delete_scheduled_taskBothRemoves a scheduled task.

Monitors

A monitor watches every incoming message on Discord, Telegram, or both. When a message matches, SILV runs your instructions automatically in a one-shot agent run — it can reply, escalate, cross-post, whatever you told it to do.

Match syntax — a case-insensitive substring (e.g. support) or a regex pattern with flags (e.g. /outage|down/i).

Optional chat filter — restrict a monitor to a single channel or chat instead of everything the bot can see.

No confirmation — automated runs act immediately without asking you first. Word the instructions carefully, and test in a quiet channel before pointing a monitor at a busy one.

Scheduled tasks

Scheduled tasks run instructions on a timer, in two flavors:

daily

Runs once a day at a fixed HH:MM in the server's local time zone. Good for digests and reports.

interval

Runs every N minutes. Good for periodic checks — remember each run costs API calls, so pick the slowest interval that works.

Both monitors and schedules work even with only one platform connected, and both show up in the dashboard's Automations sidebar with a live activity log.

Good example prompts

read

> summarize the last 100 messages in #announcements

monitors

> reply to anything in Discord that mentions 'bug' — ask for repro steps and point them to the support channel

monitors

> watch the Telegram group for /urgent messages and forward them to #alerts on Discord

schedules

> every morning at 9:00, post yesterday's highlights from #general to my Telegram channel

schedules

> every 30 minutes, check #status for error reports and ping me if anything looks broken

manage

> list my active monitors and delete the one watching for 'gm'

Limits & behavior

Up to 25 tool steps per turn Long multi-part jobs may need to be split into smaller instructions.
One chat turn at a time A busy guard queues you out while a turn is in flight — automations aren’t blocked, since they run in the background on their own transcripts.
12-hour sessions Credentials live in server memory only, scoped to your browser session, for at most 12 hours.
Everything resets on disconnect or restart Disconnecting (or a server restart) tears down the bots and wipes keys, monitors, and schedules. Re-create automations after reconnecting.

No built-in auth — SILV AI is meant for local, self-hosted use. Add authentication before exposing it on the public internet. Read the security page →

Ready to try it? Connect your keys at /console — or start from the getting started guide →