Research Agents
Research Agents are autonomous AI workers that synthesize market intelligence, analyze documents, and produce structured research reports. They operate within your organization's data boundary and require human approval before findings affect forecasts.
Agent Types
| Agent | Description | Plan |
|---|---|---|
| Market Scanner | Aggregates news, SEC filings, and market reports for competitors | Growth+ |
| Document Analyst | Extracts KPIs and trends from uploaded PDFs and spreadsheets | All |
| Macro Economist | Correlates macroeconomic indicators with your revenue drivers | Enterprise |
| Custom (LangGraph) | Build custom multi-step agent workflows with LangGraph | Enterprise |
Starting a Research Session
- Navigate to Research → New Session.
- Select an agent type and provide a research objective in plain English.
- Attach relevant context:
- Uploaded documents (PDF, DOCX, XLSX)
- Web URLs to crawl
- Dataset columns to correlate against
- Set the max steps limit (default: 20) to control cost.
- Click Start Research. The agent begins autonomously, streaming steps to the UI.
Human-in-the-Loop Review
When an agent reaches a decision point requiring external data or significant action, it pauses and requests human approval via a Review Card. You can:
- Approve — allow the agent to proceed with the proposed action
- Reject — provide alternative instructions and resume
- Terminate — stop the session immediately
All approvals are logged in the Audit Log with the actor, timestamp, and action taken.
Research Documents
Agent outputs are stored as Research Documents linked to the session. Each document includes:
- Structured findings (JSON)
- Source citations
- Confidence scores per finding
- Vector embeddings stored in ChromaDB for semantic search
Documents are searchable via Research → Document Search using natural language queries.
Integrating Findings with Forecasts
Approved research findings can be used to:
- Adjust factor values in the Knowledge Graph
- Create forecast scenarios based on market signals
- Feed document context into LLM-powered commentary on forecast dashboards
LLM Configuration
Research Agents use LLMs for reasoning. Configure the LLM provider under Settings → LLM:
- OpenAI (GPT-4o recommended)
- Anthropic (Claude 3.5 Sonnet)
- Ollama (local, self-hosted — Enterprise only)
API keys are stored encrypted in the database using AES-256 encryption.