Reference
7 invocation patterns
The seven ways a Sovereign agent can be triggered. Six produce independent threads. One — Live Mode — maintains a continuous thread. Pick the pattern that matches the trigger.
The independent thread model
Almost every invocation pattern produces an independent thread. That means: each fire starts a clean conversation, and the chat history of prior fires does not carry over. What does carry over is durable state: the agent's system prompt, its skills, its memories, its rubrics, and any connected integrations.
This is on purpose. Independent threads make automated runs predictable. If your Tuesday brief crashed on a bad input, your Wednesday brief is not contaminated by the failure. If you change a skill, every future run picks it up immediately.
Live Mode is the exception. See the bottom of this page.
The seven patterns
| Pattern | Trigger | Common use |
|---|---|---|
| Manual thread | You open the portal and start a thread | One-off work, exploration, anything you want to watch as it runs |
| Scheduled | Cron expression in the agent’s configuration | Daily morning briefing, hourly inbox triage, end-of-week roll-up |
| Webhook | An external system POSTs to a Sovereign-issued URL | CRM event drives an action. Form submission spawns a workflow. Bug tracker assigns the agent a ticket. |
| Inbound email to a Sovereign-issued address | Forward an email to the agent and have it draft a reply, file a task, run a calculation | |
| Slack | Mention in a channel, DM, or scheduled trigger in a channel context | Team-shared agent that lives in a channel and responds when called |
| Telegram | Message to the bound Telegram chat | Personal operator-facing channel. The way most Sovereign customers run their agent day to day. |
| Live Mode | Always-on monitoring of a connected channel or queue | Continuous inbox triage. Open-tickets watcher. Lead-pipeline monitor that catches a hot reply within seconds. |
Manual thread
Trigger. You open the portal and start a thread
Use it for. One-off work, exploration, anything you want to watch as it runs
State persistence. Thread persists. Agent state updates if you accept proposed skills or memories at the end.
Scheduled
Trigger. Cron expression in the agent’s configuration
Use it for. Daily morning briefing, hourly inbox triage, end-of-week roll-up
State persistence. Each fire is an independent thread. The agent carries its skills, memories, and rubrics in. The chat history of yesterday’s scheduled run does not follow into today’s.
Webhook
Trigger. An external system POSTs to a Sovereign-issued URL
Use it for. CRM event drives an action. Form submission spawns a workflow. Bug tracker assigns the agent a ticket.
State persistence. Independent thread per webhook fire. Payload becomes the opening message.
Trigger. Inbound email to a Sovereign-issued address
Use it for. Forward an email to the agent and have it draft a reply, file a task, run a calculation
State persistence. Independent thread per inbound email. Threading on subject is supported but state does not bleed between conversations.
Slack
Trigger. Mention in a channel, DM, or scheduled trigger in a channel context
Use it for. Team-shared agent that lives in a channel and responds when called
State persistence. Per-mention independent thread. The agent inherits the deploying user’s permissions — it acts as them, with their access to integrations.
Telegram
Trigger. Message to the bound Telegram chat
Use it for. Personal operator-facing channel. The way most Sovereign customers run their agent day to day.
State persistence. Same model as Slack: independent threads, deployer-scoped permissions, no chat-history leak between threads.
Live Mode
Trigger. Always-on monitoring of a connected channel or queue
Use it for. Continuous inbox triage. Open-tickets watcher. Lead-pipeline monitor that catches a hot reply within seconds.
State persistence. Exception to the independent-thread model. Live Mode maintains a single continuous thread for the lifetime of the monitor.
Picking a pattern
Three questions decide it:
- What fires the run? Time, event, message, or you.
- Who needs to be in the loop? Just you, a team channel, or a customer-facing surface.
- Does the run need to react in real time? If yes, Live Mode. If not, anything else.
Combining patterns
Agents can be triggered by multiple patterns simultaneously. The same agent might fire a scheduled run at 7am each morning, accept inbound emails to a dedicated address, and respond to Slack mentions when called. Each invocation is still its own independent thread; the patterns coexist without interfering.
Webhooks — the integration backbone
Webhook invocation is how most external systems talk to Sovereign. The pattern: external system fires a POST to a Sovereign-issued URL with a JSON payload. The payload becomes the opening message of an independent thread. The agent runs whatever it runs and produces an output the calling system can consume — either by polling a results endpoint or by receiving a callback to a URL the caller provided.
Webhooks pair naturally with our connector catalog. CRM moves a deal to a stage, webhook fires, agent does the follow-up work, agent updates the CRM. The loop closes without you in it.
Live Mode — the exception
Live Mode is the one pattern that breaks the independent-thread model. It keeps a single continuous thread running while it monitors a connected source. It is appropriate when latency matters: a hot lead reply you need to catch within seconds, a production incident you need someone watching for around the clock.
Because Live Mode runs continuously, it costs more than periodic invocations. Use it where the latency wins justify the cost. For anything that can tolerate a five-minute or hourly check, scheduled fires are cheaper and behave more predictably.