Background Agents
The specialized processes that analyze, research, and reason in parallel.
Overview
Background agents run alongside your conversation. Each is optimized for a specific type of analysis.
How it works:
- You send a message
- Each agent evaluates whether to activate
- Activated agents run in parallel (non-blocking)
- Insights surface as follow-up messages
Default Agents
Emotional Resonance
What it does: Senses the emotional undercurrent in your messages—even when you're not explicitly sharing feelings.
When it activates:
- Vulnerability or personal struggle
- Strong emotion (frustration, excitement, worry)
- Unusual brevity or tone shifts
Why it matters: Sometimes what you need isn't analysis—it's to feel heard. This agent catches emotional signals you might not voice directly.
Pattern Recognition
What it does: Spots recurring themes, behavioral loops, and contradictions across your conversation history.
When it activates:
- Same topic keeps coming up in different forms
- Contradictory statements over time
- Avoidance patterns
Why it matters: We often repeat ourselves without realizing it. This agent surfaces patterns that might reveal deeper priorities or blocks.
Confirmation Bias Detection
What it does: Flags overconfidence, unchecked assumptions, and blind spots.
When it activates:
- Absolute certainty without evidence ("This will definitely work")
- Assumptions stated as facts ("Everyone knows...")
- Dismissing alternatives without consideration
Why it matters: We're all prone to confirmation bias. This agent provides a reality check before decisions solidify.
Background Analysis
What it does: Deep analytical processing—surfaces hidden meanings, logical gaps, and implications.
When it activates:
- Explicit requests for deeper analysis
- Complex decisions with multiple factors
- Requests to "read between the lines"
Why it matters: Sometimes you need slow, deliberate thinking—not a quick take. This agent does the heavy reasoning.
Deep Research
What it does: Comprehensive multi-source research that synthesizes information across time and perspectives.
When it activates:
- Requests for thorough research
- Complex topics requiring multiple angles
- Historical context or evolution questions
Why it matters: Single web searches give surface answers. This agent builds comprehensive understanding from multiple sources.
Strategic Planning
What it does: Transforms goals into concrete action sequences with risk awareness.
When it activates:
- Requests for plans, roadmaps, or strategies
- Decisions with tradeoffs to consider
- Prioritization questions
Why it matters: Ideas need execution paths. This agent turns abstract intentions into actionable steps.
Reasoning
What it does: Mathematical, logical, and first-principles analysis.
When it activates:
- Calculations and quantitative problems
- Physics, statistics, engineering questions
- Logical derivations
Why it matters: Some problems need precise computation, not conversation. This agent handles formal reasoning.
Online Tool Use
What it does: Accesses your connected services—email, calendar, files.
When it activates:
- "Check my inbox"
- "What's on my calendar?"
- "Find that document about..."
Why it matters: Your data lives in many places. This agent retrieves and acts on it so you don't context-switch.
See Integrations for connecting services.
Triggering Agents Explicitly
You can request specific analysis:
| Say this | Triggers |
|---|---|
| "Research this thoroughly" | Deep Research |
| "Am I missing something?" | Confirmation Bias Detection |
| "Help me plan this out" | Strategic Planning |
| "Check my inbox" | Online Tool Use |
| "What patterns do you notice?" | Pattern Recognition |
Customization
You can configure:
- Which agents are active – Enable/disable specific background agents
- Which models power them – Swap in different AI providers
- Activation sensitivity – Adjust how readily agents engage
Access agent settings in your Viventium dashboard.
How Agents Choose Their Models
Each agent can use a different AI model optimized for its task:
| Agent Type | Typical Model Choice | Why |
|---|---|---|
| Fast dialogue | Grok, GPT-4o | Speed |
| Deep reasoning | Claude Opus, o1 | Accuracy |
| Research | Perplexity Sonar | Source synthesis |
| Quick classification | Llama via Groq | Ultra-low latency |
You control the mix. Experiment to find your ideal configuration.