Autothink
Autothink
Autothink gives your characters structured thinking strategies. Instead of responding immediately, the character first applies a specific reasoning methodology — deductive logic, pattern recognition, scenario planning, or one of 11 other approaches — then generates a more considered response.
The 14 Strategies
Foundation (Level 1)
| Strategy | What It Does |
|---|---|
| Deductive | Applies general principles to reach specific conclusions |
| Inductive | Observes patterns to form general rules |
| Analytical | Breaks complex problems into components (default fallback) |
| Sequential | Follows step-by-step logical progression |
| Comparative | Compares different approaches and solutions |
| Reflective | Reviews and evaluates previous decisions |
| Synthesis | Combines disparate ideas into unified solutions |
Advanced (Level 2)
| Strategy | What It Does |
|---|---|
| Critical | Questions assumptions, challenges approaches |
| Systematic | Applies structured methodologies (SWOT, 5-Why, root cause) |
| Contextual | Considers environmental factors and broader implications |
| Pattern Recognition | Identifies recurring themes and trends |
| Causal Analysis | Traces cause-and-effect chains |
| Scenario Planning | Explores multiple future possibilities |
| Meta-Cognitive | Thinks about thinking — evaluates reasoning quality itself |
How It Works
When Autothink is triggered (via strategy selection in Strings or a ThinkThing node):
- The character's personality, memories, and goals are loaded
- The selected strategy's reasoning template is applied
- The LLM performs multi-step reasoning (configurable 1-10 iterations)
- Reasoning steps and a final conclusion are produced
- The reasoning is injected into the character's response context
In Strings, you see:
- "Thinking..." indicator while reasoning runs
- Thinking metadata (strategy used, steps completed, duration, tokens consumed) after completion
- The final response informed by the reasoning
Using Autothink
In Strings
Two ways to trigger Autothink:
- ThoughtChain dropdown (0-10) — sets how many explicit reasoning steps to perform
- Autothink Strategy dropdown — selects a specific strategy
Both are in the input bar controls. You can combine them — select "Causal Analysis" strategy with 5 ThoughtChain iterations for a deep cause-and-effect analysis.
In ThinkThing
The Autothink node applies a thinking strategy to content flowing through a graph. Connect it to an Anima node and select the strategy in the node configuration.
Automatic Selection
Characters can be configured with a default strategy and weighted strategy modes. If no strategy is explicitly selected, the character uses:
- Its configured default strategy (if set)
- Weighted random selection from its strategy modes (if configured)
- Analytical (universal fallback)
Enabling Autothink
- Add the Autothink node to your character's canvas in the Visual Builder
- Configure optional defaults:
- Default strategy — used when no strategy is explicitly selected
- Strategy modes — weighted probabilities for automatic selection
Without the Autothink node, strategy selections in Strings have no effect — the character responds normally without multi-step reasoning.
Fallback Behavior
If the Autothink service is temporarily unavailable:
- The system falls back to a static reasoning prompt — a text-based instruction that tells the LLM to reason step-by-step
- The character still provides a thoughtful response, just without the full multi-step pipeline
- This fallback is seamless — users see a response, not an error
Cost and Performance
Each Autothink session makes additional LLM calls for the reasoning steps. A 3-step reasoning session = 3 extra LLM calls on top of the response generation.
| Setting | Impact |
|---|---|
| More iterations | Deeper reasoning but higher cost and latency |
| Advanced strategies | Similar cost to foundation strategies — the difference is in the prompt template, not the call count |
Tip: For casual conversation, don't use Autothink. Reserve it for questions where deeper reasoning genuinely improves the answer — complex analysis, strategic decisions, debugging, or creative problem-solving.
