🤖 Agentic Chat
Agentic Chat supercharges Curiositi by shifting from pure semantic search into interactive, continuous workflows powered by Large Language Models (LLMs). This means you aren’t just finding a single file—you can ask complex multi-part questions, and Curiositi will actively execute tools to research, extract, and synthesize the answer.
What is an Agent?
Section titled “What is an Agent?”An Agent in Curiositi is a configured AI assistant that holds a persistent context window to carry out instructions based on a system prompt. You can create multiple agents per workspace, giving each different directives or tool sets. For instance, you could have a “Legal Analyst Agent” prioritizing regulatory documents, and an “Engineering Assistant Agent” with access to codebase metadata.
Creating and Managing Agents
Section titled “Creating and Managing Agents”- System Prompts: Customize the personality and goals of the agent.
- Provider Choice: Select which underlying backend powers the agent’s logic (e.g., OpenAI, Google, Anthropic, Ollama).
- Tool Access Constraints: Define the maximum number of multi-step tool calls the Agent is allowed to perform, controlling autonomy vs reliability.
System Agents
Section titled “System Agents”Curiositi provides two built-in system agents:
| Agent | ID | Description | Max Tool Calls |
|---|---|---|---|
| Ask | system:ask | General-purpose assistant for everyday questions and tasks | 10 |
| Deep Research | system:deep-research | Thorough research agent that explores topics in depth, cites sources, and provides comprehensive analysis | 100 |
Conversations
Section titled “Conversations”Curiositi manages chat history in Conversations. This gives you persistent records of interactions. When you re-open a Conversation, the context of previous tool calls and answers remains intact, allowing you to pick up exactly where you left off.
Conversations can originate from different sources (currently web and slack), tracked via the source field.
Tools and Capabilities
Section titled “Tools and Capabilities”The true power of Agentic Chat stems from the Tools available for agents to utilize. Unlike simple chatbots, Curiositi agents can autonomously query the platform and external services.
Built-in Tools
Section titled “Built-in Tools”Curiositi agents ship with built-in primitives to understand your data:
- File Search (
fileSearch): Agent can retrieve vectorized chunk data to read direct sections of PDFs, Docs, spreadsheets, or images. Supports filtering by spaces, file types, and similarity thresholds. - Web Search (
webSearch): The agent can look up current external information using Firecrawl and blend it with your internal context. Supports domain filtering (include/exclude lists). - Web Fetch (
webFetch): Fetch and extract content from a specific URL when you have a direct link. Includes SSRF protection to prevent requests to private/internal addresses.
Model Context Protocol (MCP) Support
Section titled “Model Context Protocol (MCP) Support”Curiositi embraces the Model Context Protocol (MCP) standard. This allows you to add custom tools and context servers running anywhere (locally or hosted) simply by configuring an MCP server endpoint.
- Drop-in Extensibility: Add capabilities like Jira ticketing, GitHub issue reading, or direct database connections without writing a new plugin for Curiositi.
- Agent specific configurations: You can tie specific MCP tools to specific Agents dynamically.
- Tool Discovery: MCP servers automatically discover and register available tools when connected.
Supported AI Providers
Section titled “Supported AI Providers”| Provider | Default Model | Requires API Key |
|---|---|---|
| OpenAI | gpt-5-mini | OPENAI_API_KEY |
gemini-3-flash-preview | GOOGLE_GENERATIVE_AI_API_KEY | |
| Anthropic | claude-haiku-4-5 | ANTHROPIC_API_KEY |
| Ollama | (user-configured) | OLLAMA_BASE_URL |
Quick Actions
Section titled “Quick Actions”- Open the Chat interface from application sidebar.
- Switch between different Agents to match the problem you are solving.
- Add additional MCP server links in Workspace Settings to vastly expand Agent utility.