tool-sdk

One-liner

Let the AI invoke a first-class tool—query your ERP, call internal BOM APIs, drive lab equipment, etc. The platform injects credentials, validates parameters, and writes audit logs while you only implement the call path.

Install

Email for access

Request ragbase-tool-sdk at info@nox-lumen.com.

pip install ragbase-tool-sdk

Five-minute sample

ragbase-cli init tool my-erp-query
cd my-erp-query
vi src/my_erp_query/main.py

from ragbase.tool_sdk import Tool, tool_action
 
class ErpQueryTool(Tool):
    @tool_action(
        name="query_inventory",
        description="Look up stock by material code",
        params={"material_code": "str, material code"},
    )
    def query_inventory(self, ctx, material_code: str):
        token = ctx.credentials["erp_api_key"]   # injected at runtime
        resp = ctx.http.get(
            f"https://erp.internal/inv/{material_code}",
            headers={"Authorization": f"Bearer {token}"},
        )
        return {"qty": resp.json()["available"]}

@tool_action surfaces description / params directly to the LLM as a tool schema.

Why SDK vs “let the model write HTTP”

Concern	Raw LLM HTTP	tool-sdk
Credentials	Tokens may leak into prompts	Injected at runtime, never in LLM context
Validation	Hope the model behaves	Pydantic validation, hard failures
Auditing	None	Ledger on every call
Retries / timeouts / rate limits	DIY	Framework defaults
Tenant isolation	DIY	Platform enforced

Industry sparks

Industry	Idea
Automotive	Internal ASIL APIs; Vector vTESTstudio result fetchers
Manufacturing	ERP stock / BOM; MES job dispatch; SCADA reads
Office	Internal approval / HR / finance APIs