Hexagonal Intelligence Banner

Documentation

Settings

All settings are passed via the params dictionary. This page documents all 17 user-configurable parameters.

Python 
result = await run_analysis(
    query="Your query here",
    groq_api_key=key,
    params={
        "mode": "balanced",
        "model": "openai/gpt-oss-120b",
        "max_iterations": 3,
        # ... other settings
    }
)

Model Selection

model

Which LLM model to use for analysis.

Type String
Allowed Values "openai/gpt-oss-20b", "openai/gpt-oss-120b"
Default "openai/gpt-oss-20b"

When to change:

Choose When
gpt-oss-20b Cost-efficient analysis, faster execution, simpler queries
gpt-oss-120b Complex reasoning, research tasks, when accuracy is paramount
💡

Tip

The larger model costs roughly 2x more but can handle nuanced problems that the smaller model struggles with. Start with 20B and upgrade if results are inconsistent.

Execution Mode

mode

Pre-configured execution profile that adjusts multiple settings at once.

Type String
Allowed Values "quick", "balanced", "thorough", "research"
Default "quick"

Mode comparison:

Execution Modes Comparison

Visual Comparison of Execution Modes

Mode Iterations Timeout Best For
quick 1-2 Short Simple questions, fast responses
balanced 2-3 Medium Most general analysis tasks
thorough 3-4 Long Complex problems, important decisions
research 4-5 Extended Deep research, comprehensive analysis

When to change:

  • Use quick for straightforward factual questions or when testing
  • Use balanced for typical analysis work (good default)
  • Use thorough when the answer matters and you need confidence
  • Use research for multi-faceted problems requiring deep exploration

Iteration Control

max_iterations

Maximum number of analysis iterations before forcing synthesis.

Type Integer
Allowed Values 1, 2, 3, 4, 5
Default 2

When to change: Increase for complex problems requiring deep exploration. Decrease for simple queries to save cost.

Effect: More iterations = more thorough analysis but higher cost.

min_iterations

Minimum iterations before allowing early termination.

Type Integer
Allowed Values 1, 2, 3
Default 1

When to change: Increase when you want to ensure the system explores multiple approaches before concluding, even if early results look good.

enforce_iterations

Whether to enforce minimum iterations before allowing termination.

Type Boolean
Allowed Values True, False
Default False

When to change: Set to True when you want guaranteed multi-pass analysis regardless of intermediate quality scores.

Planning Configuration

candidate_plans

Number of candidate plans to generate before selecting the best one.

Type Integer
Allowed Values 1, 2, 3, 4, 5
Default 3

When to change:

Value Situation
1 Simple queries where strategy diversity doesn't help; saves cost
3 Default; good balance of options vs. cost
5 Complex problems where the right approach isn't obvious

enable_complexity_analysis

Whether to analyze query complexity before planning.

Type Boolean
Allowed Values True, False
Default True

Effect: Complexity analysis helps tune task count and strategy appropriately for your query.

When to change: Disable for simple queries to save one LLM call (minor cost savings).

Temperature & Creativity

temperature_offset

Adjustment to base temperatures across all operations.

Type Float
Allowed Values -0.5 to 0.5
Default 0.0

When to change:

Offset Effect Use For
Negative (e.g., -0.3) More focused, consistent, deterministic Factual queries, reproducible results
Zero (0.0) Balanced (default) Most tasks
Positive (e.g., +0.3) More creative, varied, exploratory Brainstorming, diverse perspectives
💡

Tip

When running ensemble analyses (multiple runs of the same query), try incrementing temperature offset slightly for each run to get diverse perspectives while staying within reasonable bounds.

Cost Control

cost_limit

Maximum cost allowed for the entire analysis (in USD).

Type Float
Allowed Values 0.01 to 1.00
Default Varies by mode

When to change: Adjust based on your budget and the importance of the analysis.

⚠️

Warning

Analysis terminates immediately with a FATAL error if the cost limit is exceeded. Set this higher than you expect to need to avoid premature termination.

Retry Configuration

global_run_attempts

Number of complete analysis restart attempts on failure.

Type Integer
Allowed Values 1, 2, 3
Default 1

When to change: Increase for critical analyses where you want automatic recovery from failures. Each restart is a complete re-run, so costs can multiply.

execution_retry_limit

Maximum retries when tasks fail or are rejected within an iteration.

Type Integer
Allowed Values 0 to 10
Default 3

When to change: Increase for difficult computational tasks that may need multiple attempts. Decrease if you want faster failure.

Output Configuration

save_mode

Where to save analysis outputs.

Type String
Allowed Values "local", "cloud", "none"
Default "none"

Options:

Mode Description Output Location
none No files saved Data returned in memory only
local Save to disk outputs/{agent_id}/ directory
cloud Save to S3 output/{agent_id}/ in configured bucket

focused_answer_type

Request a specific answer format in addition to the full synthesis.

Type String
Allowed Values See table below
Default "none"

Available types:

Type Output Options
none No focused answer (default)
number Numeric value
yes/no "Yes" or "No"
true/false "True" or "False"
good/bad "Good" or "Bad"
high/low "High" or "Low"
accepted/rejected "Accepted" or "Rejected"
yes/maybe/no "Yes", "Maybe", or "No"
good/neutral/bad "Good", "Neutral", or "Bad"
high/medium/low "High", "Medium", or "Low"

When to change: When you need a definitive categorical or numeric answer, especially for:

  • Automated pipelines that need predictable output formats
  • Decision support systems
  • Middleware integration

Content Safety

enable_content_filter

Whether to validate user inputs against content policy.

Type Boolean
Allowed Values True, False
Default True

Effect: When enabled, queries with inappropriate content are rejected before analysis begins.

Execution Mode

use_concurrent_execution

Whether to execute certain operations concurrently.

Type Boolean
Allowed Values True, False
Default False

When to change:

Value Situation
False Single-GPU environments; maximizes prompt caching (default)
True Multiple GPUs or high-throughput API access; faster but less cache-efficient

Advanced Settings (GPT-OSS Specific)

reasoning_effort

Controls reasoning depth for GPT-OSS models.

Type String
Allowed Values "minimal", "low", "medium", "high"
Default "low"

When to change: Increase for complex reasoning tasks that benefit from deeper model thinking. Note that higher effort means longer response times.

thinking_budget

Token budget for extended thinking (Gemini models only).

Type Integer
Allowed Values 0, -1, 1024, 2048, 4096, 8192, 16384, 24576
Default 0 (disabled)
💡

Tip

This setting is not currently applicable as GPT-OSS models don't use this feature. It's included for potential future model support.

Quick Reference

Setting Default Common Adjustments
model gpt-oss-20b Upgrade to 120b for complex tasks
mode quick Use balanced or thorough for important work
max_iterations 2 Increase to 3-4 for deep analysis
candidate_plans 3 Reduce to 1 for simple queries
temperature_offset 0.0 Negative for consistency, positive for creativity
cost_limit varies Set higher than expected to avoid cutoff
save_mode none Use local to generate HTML reports
focused_answer_type none Set when you need constrained outputs

Example Configurations

Budget-Conscious Quick Analysis

Python 
params = {
    "model": "openai/gpt-oss-20b",
    "mode": "quick",
    "candidate_plans": 1,
    "max_iterations": 1,
    "cost_limit": 0.05,
}

Thorough Research Analysis

Python 
params = {
    "model": "openai/gpt-oss-120b",
    "mode": "research",
    "candidate_plans": 5,
    "max_iterations": 5,
    "min_iterations": 2,
    "enforce_iterations": True,
    "cost_limit": 0.75,
    "save_mode": "local",
}

Automated Decision Pipeline

Python 
params = {
    "model": "openai/gpt-oss-120b",
    "mode": "thorough",
    "focused_answer_type": "yes/no",
    "temperature_offset": -0.2,  # More deterministic
    "global_run_attempts": 2,    # Retry on failure
}

Ensemble Voting (Multiple Runs)

Python 
# Run with varying temperatures
for i in range(5):
    params = {
        "mode": "balanced",
        "temperature_offset": 0.1 * i,  # 0.0, 0.1, 0.2, 0.3, 0.4
        "focused_answer_type": "yes/no",
    }
    results.append(await run_analysis(query, groq_api_key=key, params=params))