Fast Mode
Retrieve context directly from real-time data graphs without semantic search or LLM reasoning. Fast mode optimizes for low-latency interactions where agents need conversational context and stored memories without deep knowledge retrieval or cross-source reasoning.
Note: Fast mode is designed for minimal latency and doesn’t consume flex credits for context assembly.
How Fast Mode Works
When you call the GetContext API in Fast mode, Data 360 performs these operations.
- Identifies the relevant real-time data graphs for the specified individual and agent.
- Retrieves data directly from configured data graphs.
- Applies governance filters (object-level security, field-level security, record-level security) and agent-level access control.
- Returns the filtered data in a structured, text-based format optimized for LLM consumption.
Fast mode performs direct retrieval over data. It doesn’t invoke semantic search or call an LLM for relevance ranking or compression. It also doesn’t require an intent or purpose to be specified in the request.
Call GetContext in Fast Mode
To retrieve context, provide these parameters to the GetContext API.
- Conversation ID: Identifies the current conversation. Used to retrieve conversation-level data (messages, summaries).
- Owner ID: The individual or user ID of the person the agent is interacting with. Used for profile-level data lookup (facts, preferences).
- Agent ID: Your registered agent’s API name. Resolves agent configuration and determines which data graphs to query.
The API also requires a data space header and authentication credentials.
Response Format
The GetContext API returns context entries categorized by source type (such as profile or conversation). Each entry contains pre-formatted text optimized for direct injection into an LLM prompt. This includes summaries, interaction logs, and profile data in a compact notation designed for token efficiency. The response includes a unique context ID for tracing and debugging.
For the complete request parameters, headers, response schema, and sample payloads, see Get Context.
Data Sources
Fast mode retrieves data from these real-time data graphs.
| Source | Data Provided |
|---|---|
| Profile data graph | Profile-level memories (extracted facts and preferences) and summary information linked to the individual. |
| Conversation data graph | Conversation-level data including recent messages, session summaries, and interaction history for the specified conversation. |
| Additional real-time data graphs | If you configure additional data graphs during agent registration via contextDataSources, the system includes data from them. |
Fast mode does not access these sources.
- Enterprise search (Salesforce search, vector search)
- External structured data not in a configured data graph
- Knowledge graphs or world knowledge sources
- Any LLM-powered reasoning or relevance ranking
When to Use Fast Mode
Fast mode is appropriate for agents that handle straightforward interactions.
- Web chat agents providing quick answers to common questions.
- Generic customer support agents handling simple inquiries (order status, hours of operation, basic troubleshooting).
- FAQ and self-service agents where the conversation is short and focused.
- Triage agents that route customers to specialized agents after initial classification.
- Notification agents that need basic user context for message personalization.
Governance in Fast Mode
Fast mode applies the same governance model as all Agentic Memory and Context operations.
- Object-level security: If the calling user doesn’t have access to a DMO, the system excludes data from that DMO from the response.
- Field-level security: Individual fields within DMOs are redacted if the user lacks field-level access.
- Record-level security: The system includes only records the user is authorized to see.
- Agent-level access control: Each agent can only read its own data. The response excludes memories extracted by other agents.
See Also
j