P(correct tool) = P(correct family) × P(correct tool | correct family)

1.5B model: P(family)=56%, P(tool|family)=89%  →  50% overall
35B model:  P(family)=90%, P(tool|family)=98%  →  88% overall

---

from langchain_core.tools import tool

@tool
def file_read(path: str, encoding: str = "utf-8", 
              line_numbers: bool = False, offset: int = 0) -> str:
    """Read file contents with line numbers, offset, and encoding control."""
    with open(path, encoding=encoding) as f:
        content = f.read()
    if line_numbers:
        lines = content.split("\n")
        content = "\n".join(f"{i+1}: {l}" for i, l in enumerate(lines))
    if offset:
        content = "\n".join(content.split("\n")[offset:])
    return content

---

from langchain_core.tools import tool

@tool(
    tier_descriptions={
        "small":  "Read file",
        "medium": "Read a file from disk",
        "large":  "Read file contents with line numbers, offset, and encoding control",
    },
    tier_params={
        "small":  ["path"],
        "medium": ["path", "encoding"],
        "large":  ["path", "encoding", "line_numbers", "offset"],
    },
    category="filesystem",
)
def file_read(path: str, encoding: str = "utf-8",
              line_numbers: bool = False, offset: int = 0) -> str:
    """Read file contents with line numbers, offset, and encoding control."""
    with open(path, encoding=encoding) as f:
        content = f.read()
    if line_numbers:
        lines = content.split("\n")
        content = "\n".join(f"{i+1}: {l}" for i, l in enumerate(lines))
    if offset:
        content = "\n".join(content.split("\n")[offset:])
    return content

---

from langchain_openai import ChatOpenAI
from langchain.agents import create_tool_calling_agent

# Small local model
llm_small = ChatOpenAI(
    base_url="http://localhost:11434/v1",
    model="qwen2.5:1.5b",
)

# Tools automatically adapt: small model gets 
# "Read file" with only `path` parameter
agent = create_tool_calling_agent(
    llm=llm_small,
    tools=[file_read, web_search, git_status, ...],
    model_tier="small",  # or auto-detect from model name
)

---

# Tier "small" (1.5B) — what the model receives:
{
    "name": "file_read",
    "description": "Read file",
    "parameters": {
        "type": "object",
        "properties": {"path": {"type": "string"}},
        "required": ["path"]
    }
}

# Tier "large" (35B+) — what the model receives:
{
    "name": "file_read",
    "description": "Read file contents with line numbers, offset, and encoding control",
    "parameters": {
        "type": "object",
        "properties": {
            "path": {"type": "string"},
            "encoding": {"type": "string"},
            "line_numbers": {"type": "boolean"},
            "offset": {"type": "integer"}
        },
        "required": ["path"]
    }
}

---

from langchain.agents import TierRouter  # proposed

router = TierRouter(
    tools=[file_read, web_search, grep, git_status, ...],
    model_tier="small",
    strategy="hybrid",  # 8 detailed + rest name-only
)

# For each user query, router selects and adapts tools
adapted_tools = router.route("Read the file config.yaml")
# Returns [file_read, file_write, ...] with small-tier schemas