Skip to main content

tool-prune

Calibrated tool selection and schema pruning for AI agents. Zero-dependency offline TurboQuant, custom local decision engines, or TypeSafe System One.

pip install tool-prune

# Optional Rust SIMD acceleration (0.041 ms search)
pip install turbovec numpy

# Optional: TypeSafe API key for cloud reasoning
export TYPESAFE_API_KEY="apikey_..."

Quick start

from tool_prune import prune

tools = {
    "read_file": "Read raw text from local filesystem path",
    "run_query": "Execute SQL queries against database",
    "web_search": "Search public web for documentation or articles"
}

match = prune("what tables exist in the db?", tools)
print(match.tool)       # 'run_query'
print(match.confidence) # 0.815
print(match.engine)     # 'turboquant'

prune() runs offline via 4-bit PolarQuant + 1024-bit QJL TurboQuant (1.35 ms pure stdlib, 0.041 ms with turbovec, 100% BFCL v3 Top-1 and 100-Tool Top-5 recall) by default, or routes to TypeSafe System One when TYPESAFE_API_KEY is set. Accepts plain dicts, MCP inputSchema, Anthropic input_schema, and OpenAI { "type": "function", "function": ... } tool lists.

Schema pruning for LLMs

from tool_prune import prune

router = prune(mcp_tools)

# Conformal auto-selection (or pass k=5):
top_tools = router.filter(user_prompt, allow_empty=True)

response = llm.chat(
    tools=top_tools,
    messages=[{"role": "user", "content": user_prompt}]
)

Cuts prompt schema tokens by up to 98% while preserving your exact SDK tool dicts.

Fast-path direct dispatch

result = router.dispatch("read ./pyproject.toml", {
    "read_file": lambda q: open("pyproject.toml").read(),
    "run_query": lambda q: db.query(q),
    "fallback": lambda q, m: call_llm(q, m.auto_tools)
})

Pluggable engines

local_match = prune(query, tools, engine="turboquant")
simd_match  = prune(query, tools, engine="turbovec")
cloud_match = prune(query, tools, engine="typesafe", api_key="...")
custom_match = prune(query, tools, engine=decision_engine)
  • turboquant: 100% offline, zero network, zero dependencies (1.35 ms pure Python stdlib).
  • turbovec: Optional Rust SIMD backend (0.041 ms kernel latency) sharing the same Corpus-IDF feature pipeline.
  • custom engine: Any object with choice({ "state", "question", "options" }) or callable (query, criteria, kwargs) -> candidates.
  • typesafe: Cloud System One reasoning (Jev) with calibrated probabilities.

Demo

python examples/quickstart.py

License

MIT © Hemanth.HM

Metadata

Release files for tool-prune 0.4.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for tool-prune 0.4.0
File Size Uploaded Provenance Actions
tool_prune-0.4.0.tar.gz 15.1 kB Without provenance Details for tool_prune-0.4.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for tool-prune 0.4.0
File Interpreter ABI Platform Actions
tool_prune-0.4.0-py3-none-any.whl Python 3 none any Details for tool_prune-0.4.0-py3-none-any.whl

Total release size: 29.9 kB

Release files / tool_prune-0.4.0.tar.gz

Download URL tool_prune-0.4.0.tar.gz
Size 15.1 kB
Tags Source
SHA-256 checksum
How to use checksums
4d9808c6b4a1d13745e197d53f01c8d810ebc34c969cd3fa94e60e498a938cdb
BLAKE2b-256 checksum
How to use checksums
915e0467d71ceb20303998b6a80f79c2a325ffaa331bcd4b30d93f108a21dc15
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.13.14

Release files / tool_prune-0.4.0-py3-none-any.whl

Download URL tool_prune-0.4.0-py3-none-any.whl
Size 14.9 kB
Tags Python 3
SHA-256 checksum
How to use checksums
a4b4e30280a10027335945de1f0874c5a871c2a0b7467f7f7357457a61080107
BLAKE2b-256 checksum
How to use checksums
4586f58f267444d0916fba8b03fc746b81c45c4baeae1d622c2c6eeb32db7bc1
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.13.14

Security for tool-prune 0.4.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page