OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Needle

Needle is an Apache-2.0 compact foundation model for local tool calling, structured extraction and text embeddings on resource-constrained devices. Its model is provided as a single 8–29 MB binary using 2-bit representation, with deployment targets including phones, wearables, robots, smart homes, vehicles and microcontrollers.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Needle — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Needle when an application needs local inference for tool selection, schema-constrained extraction or embeddings and must fit a compact model into an edge deployment.

Who it is for

  • Developers building offline or edge assistants
  • Teams exposing device functions through structured tools
  • Applications that need typed extraction from messy text
  • Edge systems requiring local embeddings for search, matching or routing

Documented uses

  • Select application-provided functions from natural-language requests
  • Fill tool arguments from user input
  • Extract typed fields according to a declared schema
  • Create text embeddings locally
  • Combine local speech transcription with downstream tool calls using the shared engine
Before you choose

Requirements & limits

The repository is public and specifies an Apache-2.0 license. The documented Python installation command is `pip install cactus-needle`.

  • The model is compact: the repository describes an 8–29 MB binary using 2-bit representation.
  • Tool responses are constrained to exposed functions; unsupported requests return an empty list rather than a guessed tool call.
  • Structured extraction uses declared schemas and a byte-level grammar designed to keep output parseable.
  • Needle supports mobile and self-hosted deployment targets.
  • The Python example returns function calls, reasoning and a calibrated confidence score.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Set up Needle for a Python tool-calling workflow

Install the Python package, expose a typed application function and pass a user request to Needle.

Before you start

  • Python environment
  • The cactus-needle package
  • A function with typed arguments and a descriptive docstring

Steps

  1. Install the package with the command below.

    Documented command
    pip install cactus-needle

  2. Decorate an application function with needle.tool.

  3. Create needle.Needle with the tool in its tools list.

  4. Call agent.run() with a natural-language request and read the returned results.

Expected outcome

The documented example returns a JSON-compatible result from the application function selected by Needle.

Guide limitations

  • The example covers tool definition and invocation, but does not document a complete device integration.
  • Needle returns an empty list when a request has no matching exposed tool.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
2 October 2026
Original website title
cactus-compute/needle · Official homepage title · retained in its source language
How we review sourcesSuggest a correction