Agents
Learn about VectorChat agents — configurable AI assistants that combine models, tools, knowledge, and skills to automate work.
Agents are the core building block of VectorChat. Each agent is a configurable AI assistant that can hold conversations, use tools, search knowledge bases, and execute skills on your behalf.
Screenshot: The agents overview page showing all configured agents with their names, models, and status.
What is an agent?
An agent combines several components into a single, purpose-built AI assistant:
- AI Model — The language model that powers the agent's reasoning (e.g., GPT-4o, Claude, Gemini).
- System Instructions — A custom prompt that defines the agent's behavior, personality, and constraints.
- Tools — Integrations with external services like Slack, Gmail, and Google Calendar that the agent can use during conversations.
- Knowledge Bases — Collections of documents the agent can search for accurate, grounded answers.
- Skills — Reusable instruction sets that teach the agent how to handle specific tasks.
Key capabilities
Conversations
Chat with your agent in a real-time streaming interface. The agent maintains conversation history within a session and can reference earlier messages for context. You can have multiple conversations with the same agent, each tracked separately.
Tool use
When an agent has tools connected, it can take actions on your behalf — sending emails, posting Slack messages, creating calendar events, and more. Tools can be configured in shared mode (organization-wide credentials) or personal mode (per-user authentication).
Document search
Agents connected to knowledge bases can search uploaded documents to provide accurate, source-backed answers.
Memory
When memory is enabled, agents can remember preferences, facts, and decisions across conversations.
Approval workflows
For sensitive actions, agents can request approval before executing a tool call. You review the proposed action and approve or deny it directly in the conversation.
Agent settings
| Setting | Description |
|---|---|
| Model | The AI model used for generating responses. |
| Temperature | Controls response creativity (0 = deterministic, higher = more varied). |
| Max Tokens | Limits the length of each response. |
| Agent Mode | Enables multi-step reasoning with tool calls. |
| Autonomy Level | Controls how independently the agent acts. |
| Think Tool | Enables a visible reasoning step before the agent responds. |
| Memory | Allows the agent to remember information across sessions. |
| Save Messages | Persists conversation history for future reference. |
| Public Access | Allows anonymous users to chat via an embeddable widget. |