Auto Tool Inventory

How `–auto` picks the tools it shows the planner

The --auto planner sees whatever the ToolResolver sees — install more tools and they show up automatically.

from praisonaiagents import Agent

# The --auto planner sees whatever the ToolResolver sees. Install more
# tools and they show up automatically — no config change.
agent = Agent(name="planner", instructions="Plan an --auto run.")
agent.start("Research current AI trends and write a summary.")

Quick Start

1. Run auto mode

   Run --auto with your OpenAI key set:

   export OPENAI_API_KEY=sk-...
   praisonai --auto "research AI trends and write a summary"

   The planner’s Available Tools: list reflects whatever is installed right now.

2. Install a tools package

   Add more tools and the planner sees them next run — no config change:

   pip install "praisonai-tools"

3. Rerun and watch the list grow

   Run the same command again — the planner now advertises the newly installed tools:

   praisonai --auto "research AI trends and write a summary"

   Confirm what the resolver can see with:

   praisonai tools list

How It Works

The planner prompt’s Available Tools: line is filled from the canonical ToolResolver, so it always matches what your agents can call at run time.

The recommended set drawn from this inventory is also the task-scoped allow-list the server-side gate enforces on the LLM’s output — any tool the LLM invents outside that set is dropped from the saved YAML with a WARNING. See Auto-Generator Safety.

If the resolver is unavailable (import failure or air-gapped environment), the planner falls back to a frozen list so --auto still produces a working plan.

Configuration Options

The inventory is driven by the resolver — configure it through the ToolResolver API rather than a separate setting.

Common Patterns

Local tools.py picked up by the planner:

export PRAISONAI_ALLOW_LOCAL_TOOLS=true
# Drop a tools.py in your working directory, then:
praisonai --auto "summarise today's news"

Register a wrapper ToolRegistry function so the planner sees it:

from praisonai import PraisonAI

def my_search(query: str) -> str:
    """Custom search function."""
    return f"results for {query}"

praison = PraisonAI(agent_file="agents.yaml")
praison.agents_generator.tool_registry.register_function("my_search", my_search)
praison.run()

Inject an explicit resolver for tests or multi-tenant runtimes:

from praisonai.agents_generator import AgentsGenerator
from praisonai.tool_resolver import ToolResolver

my_resolver = ToolResolver(tools_py_path="tools.py")

gen = AgentsGenerator(
    agent_file="agents.yaml",
    framework="praisonai",
    config_list=[{"model": "gpt-4o", "api_key": "sk-..."}],
    tool_resolver=my_resolver,
)

Best Practices

pip install "praisonai-tools"

from praisonai.tool_resolver import ToolResolver

resolver = ToolResolver(tools_py_path="tools.py")

from praisonai.tool_resolver import reset_default_resolver

reset_default_resolver()

Related