Tier — Adaptive Tool Routing for AI Agents

The Problem

Every AI agent framework shows all tools identically to every model. A 1.5B model on a Raspberry Pi gets the same 80 tool descriptions as GPT-4.

35B

88%

1.5B

50%

The model isn't bad at using tools — it's bad at finding them.

The Insight

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

Tool selection decomposes into two stages. The results are surprising:

56%

1.5B finds the right family

vs 90% for 35B

89%

1.5B picks the right tool
within the right family

vs 98% for 35B

Even a 1.5B model picks the right tool 89% of the time — when shown the right neighborhood.

The Solution

Different model sizes get different tool presentations. Same 80 tools, adapted per capability:

Tiny

< 4B parameters

8 detailed + 72 name-only

"Read file"
path: string

60%

+10pp vs baseline

Large

14B - 35B

All 80, reordered + hint

"Read file with
encoding control"
path, encoding, lines

88%

+8pp vs baseline

XL

35B+

All 80, full descriptions

"Read file with line
numbers, offset,
and encoding"
path, encoding, lines,
offset, limit

88%

baseline

Benchmark Results

1,000+ native tool calling inference calls. Ollama /api/chat with tools parameter. 80 tools, 50 prompts, 4 models.

Strategy	1.5B	9B	20B	35B	Tokens
Baseline (all 80)	50%	80%	80%	88%	2,100-5,300
Hybrid (8+72)	60%	-	76%	-	~1,800
Reorder + hint	54%	-	88%	-	same
Semantic top-8	64%	72%	70%	76%	310-724
Family oracle	70%	86%	84%	88%	400-900

Quick Start

# Install
pip install yantrikos-sdk

# Define a tier-aware tool
from yantrikos import BaseTool, ToolResult, Tier, register

@register
class FileReadTool(BaseTool):
    name = "file_read"
    category = "filesystem"

    descriptions = {
        Tier.S:  "Read file",
        Tier.M:  "Read a file from disk",
        Tier.L:  "Read file with encoding control",
        Tier.XL: "Read file with line numbers, offset, encoding",
    }

    parameters = {
        Tier.S:  {"path": str},
        Tier.M:  {"path": str, "encoding": str},
        Tier.L:  {"path": str, "encoding": str, "lines": bool},
        Tier.XL: {"path": str, "encoding": str, "lines": bool, "offset": int},
    }

# Route by model tier
from yantrikos import TierRouter

router = TierRouter(model_name="qwen2.5:1.5b")
tools = router.route("Read the file config.yaml")
# Returns native tool definitions adapted for 1.5B