How agents work
Konvio agents are AI assistants with tools — structured actions they can take on your behalf during conversations and automations.
The agent loop
- You send a message (or an automation trigger fires)
- The agent plans using its instructions and available tools
- Tools run — read repos, query logs, list docs, etc.
- The agent responds with findings, proposals, or questions
- You steer — approve, clarify, or ask for more depth
Agents do not run unattended forever in a conversation; you remain in the loop unless an automation defines a full workflow.
Tools
Tools are capabilities exposed to agents, such as:
- Listing and reading files in connected repositories
- Searching documentation collections
- Calling MCP connectors (logs, metrics, APIs)
- Running commands in configured environments (automations)
Your workspace admin controls which tools each agent can use.
Context
Agents receive context from:
- The selected application
- Conversation history in the thread
- Workspace configuration (agents, variables, connectors)
- Tool results from the current turn
Be explicit in prompts — agents do not automatically know about incidents you have not described.
Limitations
- Agents can be wrong. Verify critical conclusions against primary sources.
- Tool access is bounded by connectors and permissions you configure.
- Long threads may summarize earlier context; restate key facts when resuming old conversations.
Autonomous automations
Automations run predefined agent steps without manual messages. Design them with clear prompts and narrow tool access for production safety.