AI Agents
Intelligent agents that work alongside your apps
Build conversational AI assistants and task agents that understand your data, connect to your integrations, and take action. No API keys. No infrastructure to manage. Just describe what you need.
Handles incoming support tickets, categorizes issues, suggests solutions from your knowledge base
Scores incoming leads against your criteria, routes to the right team member automatically
Creates weekly reports from your data sources, formats them, and sends summaries to Slack
Enriches contact data from public sources, fills in missing fields, and updates your CRM records
Powered by Runwork AI Access from anywhere
Web Dashboard
Browse all agents, monitor executions in real time, view event streams with tool calls and costs, and retry failed runs from the Agents Dashboard.
Open in dashboardCLI
Capabilities
Chat Agents
Build conversational AI assistants with persistent memory. Chatbots, data analysts, onboarding guides. Any chat-based experience.
Sandbox Delegation
Agents delegate complex tasks to their own computing environment. Full file system, web access, and workspace tools. "Go analyze this spreadsheet and come back with a report."
Task Agents
Create automation agents that execute multi-step tasks. Data processing, report generation, workflow orchestration.
Integration Access
Agents automatically get tools for connected integrations. Query HubSpot, create GitHub issues, send Slack messages. All through chat.
Entity Access
Agents can read and write your workspace data. Ask questions about your data or let agents update records on your behalf.
Custom Tools
Define app-specific tools with Zod schemas. Calculate metrics, trigger workflows, call external APIs. Add any capability you need.
Memory Strategies
Choose how agents remember: session (resets each conversation), user (per-user persistence), or shared (global knowledge base).
Agents Dashboard
A dedicated page to browse all registered and ad-hoc agents. See execution history with status, duration, cost, and token usage. Watch live execution streams and retry failed tasks.
Execution Monitoring
Watch agents work in real time. Each execution shows tool calls, text output, and errors as they happen. Token usage, cost per run, and triggered-by user tracked automatically.
Use Cases
Why It Matters
- No API keys or LLM infrastructure to manage
- Agents are defined in code but generated via chat
- Full access to workspace data and integrations
- Built-in streaming, memory, and tool calling
How It Works
AI Agents in Runwork are intelligent assistants that combine the power of large language models with direct access to your business data and tools. Unlike traditional chatbots that can only answer questions, Runwork agents can actually take action—querying databases, updating records, triggering automations, and connecting to external services. With sandbox delegation, agents can even spin up their own computing environment for complex tasks like data analysis, file processing, and multi-step research.
Creating an agent starts with a simple conversation. Tell the Work Assistant what you want your agent to do: "Create a customer support bot that can look up order status and process refunds." Runwork generates the agent code, complete with the necessary tools and memory configuration. You can then customize behavior, add guardrails, or expand capabilities through natural language.
Every agent automatically inherits access to your workspace. This means they can query your entities (customers, orders, products), call your connected integrations (Slack, HubSpot, Stripe), and trigger your workflows. There's no manual wiring—agents understand your data model and available actions out of the box. From any AI tool connected to your workspace, you can list agents, check their status, and review execution history. The CLI gives you runwork agents list and runwork agents executions for direct access.
Memory is handled automatically based on your chosen strategy. Session memory keeps context within a single conversation. User memory persists information across sessions for each individual user—perfect for remembering preferences or past interactions. Shared memory creates a global knowledge base that all users can contribute to and query from.
Agents run on Runwork's infrastructure with built-in streaming for real-time responses, automatic retry handling, and usage tracking. You never manage API keys, rate limits, or model deployments. Just describe what you need, and your agent is ready to help your team.
The Agents Dashboard gives you full visibility. Browse all registered agents and ad-hoc executions in one place. Each execution shows its status, duration, cost, and token usage. Expand any run to see the full event stream: every tool call, text response, and error in order. If something fails, retry with one click. The dashboard auto-refreshes every 10 seconds while agents are running, so you always see the latest state. From the CLI, runwork agents executions gives you the same view, and runwork agents execution <sessionId> shows the full stream for any run.