Praison AI

Agent Execution

Control agent execution limits – iterations, time, rate limiting, and retries

See also: Five-Layer Agent Stack — Execution is Layer 4 (the Loop).

Execution configuration controls how agents run – maximum iterations, time limits, rate limiting, and retry behavior.

Diagram

Execution Limits

🔄 Iterations
max_iter=20

⏱️ Time
max_execution_time

📊 Rate
max_rpm

🔁 Retries
max_retry_limit

🤖 Agent

Quick Start

  1. Preset Mode
    python
    from praisonaiagents import Agent

# Use a preset
agent = Agent(
    name="Fast Agent",
    instructions="You answer quickly",
    execution="fast"  # Preset: limited iterations
)
  2. Custom Configuration
    python
    from praisonaiagents import Agent, ExecutionConfig

agent = Agent(
    name="Thorough Agent",
    instructions="You do comprehensive analysis",
    execution=ExecutionConfig(
        max_iter=50,              # Max iterations
        max_rpm=100,              # Requests per minute
        max_execution_time=300,   # 5 minute timeout
        max_retry_limit=5,        # Retries on failure
    )
)

Execution Presets

Diagram

Presets

Quick tasks

Most tasks

Complex tasks

Long-running

⚡ fast
max_iter=10

⚖️ balanced
max_iter=20

🔍 thorough
max_iter=50

♾️ unlimited
No limits

Use Case

Presetmax_iterUse Case
fast10Quick responses
balanced20Standard tasks
thorough50Complex analysis
unlimitedNoneLong-running jobs
python
agent = Agent(instructions="...", execution="fast")
agent = Agent(instructions="...", execution="balanced")
agent = Agent(instructions="...", execution="thorough")
agent = Agent(instructions="...", execution="unlimited")

Configuration Options

python
from praisonaiagents import ExecutionConfig

config = ExecutionConfig(
    max_iter=20,              # Maximum iterations
    max_rpm=None,             # Requests per minute (None = no limit)
    max_execution_time=None,  # Timeout in seconds (None = no limit)
    max_retry_limit=2,        # Retries on failure
)
OptionTypeDefaultDescription
max_iterint20Maximum loop iterations
max_rpmintNoneRate limit (requests/minute)
max_execution_timeintNoneTimeout in seconds
max_retry_limitint2Max retries on failure

Iteration Limits

Prevent infinite loops:

Diagram
CounterAgentCounterAgentalt[count >=max_iter]loop[Until done or max_iter]Execute stepIncrementCountStop
python
agent = Agent(
    instructions="You solve problems iteratively",
    execution=ExecutionConfig(max_iter=30)
)

# Agent stops after 30 iterations even if not done

Time Limits

Prevent runaway execution:

python
agent = Agent(
    instructions="You analyze data",
    execution=ExecutionConfig(
        max_execution_time=60  # Stop after 60 seconds
    )
)
Diagram

Time < limit

Time >= limit

▶ Start

⏱️ Timer

Execute

Continue

⏹️ Stop

Rate Limiting

Control API request rate:

python
agent = Agent(
    instructions="You search the web",
    execution=ExecutionConfig(
        max_rpm=60  # Max 60 requests per minute
    )
)

Useful for:

  • Respecting API rate limits
  • Controlling costs
  • Preventing abuse

Retry Behavior

Handle transient failures:

python
agent = Agent(
    instructions="You call external APIs",
    execution=ExecutionConfig(
        max_retry_limit=3  # Retry up to 3 times
    )
)
Diagram
APIAgentAPIAgentRetry 1Retry 2RequestErrorRequestErrorRequestSuccess

Multi-Agent Execution

For multi-agent workflows:

python
from praisonaiagents import AgentTeam, MultiAgentExecutionConfig

team = AgentTeam(
    agents=[agent1, agent2],
    execution=MultiAgentExecutionConfig(
        max_iter=10,      # Per task
        max_retries=5,    # Per task
    )
)

Combining Limits

Use multiple limits together:

python
agent = Agent(
    instructions="You do comprehensive research",
    execution=ExecutionConfig(
        max_iter=100,             # Allow many iterations
        max_execution_time=600,   # But stop after 10 minutes
        max_rpm=30,               # Rate limit API calls
        max_retry_limit=3,        # Retry failures
    )
)

Best Practices

Set appropriate iteration limits

Too low = incomplete tasks. Too high = wasted resources. Start with defaults and adjust.

Use time limits for production

Always set max_execution_time in production to prevent runaway agents.

Rate limit external API calls

Use max_rpm when agents call external APIs to respect rate limits.

Enable retries for unreliable operations

Set max_retry_limit for operations that may fail transiently.

Related

Output

Display configuration

 Autonomy

Independence control