Thinking Budgets

Configure extended thinking token budgets

The thinking command manages thinking token budgets for complex reasoning.

Invocations of praisonai thinking … shown here reach their handler as of PraisonAI PR #4329 (release ≥ the merge date, 2026-08-24). Older releases silently billed this exact command to an LLM at ~2.5k tokens with exit 0 — see issue #4327. Upgrade with pip install -U praisonai.

Quick Start

# Show current thinking budget
praisonai thinking status

Usage

Show Status

praisonai thinking status

Expected Output:

╭─ Thinking Budget ────────────────────────────────────────────────────────────╮
│  Level: medium                                                               │
│  Max Tokens: 8,000                                                           │
│  Adaptive: enabled                                                           │
╰──────────────────────────────────────────────────────────────────────────────╯

Set Budget Level

praisonai thinking set high

Available levels: minimal, low, medium, high, maximum

Use from code / run

For a single invocation without persisting a global default, pass --thinking on praisonai code or praisonai run:

praisonai code --thinking high "Refactor src/utils.py for readability"
praisonai run --thinking medium "Plan a release checklist"

The --thinking flag is applied after the agent is constructed. It takes effect on the very next request whether the LLM instance has been materialised yet or not (fixed in PR #4479).

Surface	Levels	Persisted?
--thinking on code / run	off, minimal, low, medium, high	No — per invocation
praisonai thinking set	minimal, low, medium, high, maximum	Yes — .praison/thinking.json

Unknown --thinking values fail closed with exit code 1 before any work runs.

Show Usage Stats

praisonai thinking stats

Python API

from praisonaiagents.thinking import ThinkingBudget, ThinkingTracker

# Use predefined levels
budget = ThinkingBudget.high()  # 16,000 tokens

# Track usage
tracker = ThinkingTracker()
session = tracker.start_session(budget_tokens=16000)
tracker.end_session(session, tokens_used=12000)

summary = tracker.get_summary()
print(f"Utilization: {summary['average_utilization']:.1%}")

See Also

• Thinking Budgets Feature
• Reasoning Effort — Session Persistence — why --thinking restores automatically on --continue