# How to run your agent from anywhere

Source: https://www.hyperagent.com/docs/lessons/how-to-run-your-agent-from-anywhere

Learn the invocation patterns that let your agent run without you: scheduled triggers, Live Mode, webhooks, Slack, Telegram, and email.

Every invocation creates an independent thread. The agent carries its skills, memories, and system prompt into each run, but conversation history doesn't follow. Understand this model first, then choose the invocation pattern that matches your trigger.

## Core concepts [#core-concepts]

### The Independent Thread Model [#the-independent-thread-model]

With the exception of Live Mode, every time you run an agent it creates an independent thread. The agent brings its skills, memories, and system prompt into each new thread automatically. Conversation history from previous threads doesn't carry over.

An agent keeps only what it learned on Friday via Skills and Memories. A Slack-triggered thread starts fresh regardless of previous interactions. Configuration carries over; conversations in threads don't.

This is the same behavior you experience in manual threads. The difference is that automated triggers can make the fresh start feel surprising if you expect continuity between runs.

### Making Your Agent Run Without You [#making-your-agent-run-without-you]

**Scheduled invocations** set a time-based trigger: every Monday at 7 AM, or the first of each month. Your prompt must be entirely self-contained, because no one will be there to clarify mid-run.

**Live Mode** keeps your agent always on, monitoring a channel and responding when triggered. This is the one exception to independent threads: Live Mode maintains a continuous conversation.

**Webhooks** react to external events. When something changes in another system, it sends that event data to your agent, which acts on it automatically.

### Running Your Agent From Outside Hyperagent [#running-your-agent-from-outside-hyperagent]

**Slack deployment** lets your team interact with the agent directly in a channel. There are several configuration options covered in a dedicated video.

**Telegram** lets you reach your agent through messaging for mobile-first or cross-platform access.

**Email** means you can send your agent an email and it responds like any other teammate, requiring no special interface.

Most agents start with manual threads during the shaping phase, then graduate to one or more of these patterns once the workflow is proven. Some use multiple simultaneously: a weekly reporting agent might run on a Monday schedule and also be available in Slack for ad-hoc questions.

## Seven ways to run your agent [#seven-ways-to-run-your-agent]

| Pattern              | Trigger                    | Thread Behavior               | Best For                            |
| -------------------- | -------------------------- | ----------------------------- | ----------------------------------- |
| Manual thread        | You open the app           | New independent thread        | Shaping workflows, ad-hoc tasks     |
| Scheduled invocation | Time-based (cron)          | New independent thread        | Weekly reports, recurring audits    |
| Live Mode            | Channel activity           | Continuous thread (exception) | Always-on monitoring, triage        |
| Webhook              | External event             | New independent thread        | System-driven reactions, alerts     |
| Slack                | Channel message or mention | New independent thread        | Team access, collaborative Q\&A     |
| Telegram             | Telegram message           | New independent thread        | Mobile-first, cross-platform        |
| Email                | Incoming email             | New independent thread        | No-interface access, external users |

## Key vocabulary [#key-vocabulary]

The act of triggering an agent to start working. Each invocation type corresponds to a different trigger source: time, event, message, or manual action.

A fresh conversation created for each agent run. The agent brings its configuration but not conversation history from prior threads.

An always-on invocation pattern where the agent monitors a channel continuously. The one exception to the independent thread model: it maintains a single continuous thread.

An event-driven trigger where an external system sends data to your agent when something changes, prompting the agent to act automatically.

A prompt that includes all necessary context and instructions for the agent to complete its work without human clarification during the run.

## FAQs [#faqs]

Each automated invocation creates a fresh thread. The agent carries its skills, memories, and system prompt into every new thread, but conversation history from previous threads doesn't follow. The agent starts each run with a clean slate of conversation context while retaining everything it's learned through the learning system. Skills and memories persist across all runs; only the in-thread conversation resets.

Live Mode is the one exception to the independent thread model. Instead of creating a new thread for each interaction, Live Mode keeps a continuous thread where the agent monitors a channel and responds when triggered. Conversation context accumulates over time within the same Live Mode session, making it suitable for ongoing monitoring and triage where conversational continuity matters.

Most agents start with manual threads during the shaping phase while you iterate on the prompt and workflow. Automation comes after the output is consistent and reliable. A good signal: when you find yourself opening the same thread, pasting the same prompt, and getting results you trust without needing to clarify or adjust mid-run. At that point, the workflow is ready for a scheduled invocation, webhook trigger, or channel deployment.

A single agent can be invoked through multiple patterns at the same time. A weekly reporting agent might run on a Monday morning schedule for its standard report and also be available in Slack for ad-hoc questions throughout the week. Each invocation still follows the independent thread model (except Live Mode), so the scheduled run and the Slack interaction create separate threads, each starting fresh with the agent's full configuration.

## Summary [#summary]

Pick one workflow you run manually every week. Choose the invocation type that fits the trigger, write a self-contained prompt, and run it as a manual thread first. When the output is consistent, automate it. The goal is an agent that runs without you.