SkillFoxx

Ollaya skill

Ollaya skill and MCP server: an agent puts typed questions to local decision models and gets probabilities with no LLM call

  • 1,459 ★ on the repository
  • Last commit
  • Apache-2.0
Skill

Extended access

Why this level

  • Runs a local daemon with an HTTP API on port 11435 and downloads model weights from Hugging Face
  • The `curl | sh` installer runs with your privileges
  • Low-confidence decisions can be wrong when the agent acts on them automatically
All reasons and checks

ollaya-dev/ollayaAuthor:Mert Cobanov

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add skills/ollaya

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Extended accessThe tool runs code or makes network calls and only reads external data.

  • Runs a local daemon with an HTTP API on port 11435 and downloads model weights from Hugging Face
  • The `curl | sh` installer runs with your privileges
  • Low-confidence decisions can be wrong when the agent acts on them automatically

This is third-party code. Review the repository files before installing.

What it does

Ollaya runs decision models on your computer. A model receives a state (an email, a ticket, any JSON) and typed questions of three kinds: `choice` picks from options, `score` rates on a scale, `noul` checks a yes or no statement. In one pass it returns probabilities and generates no text. The repository ships two artifacts for agents. The `ollaya-decisions` skill teaches an agent when to use such decisions: sorting tickets and emails, routing requests, moderation, screening prompts for jailbreaks and injections. The agent picks a path in order: the `decide` MCP tool first, then the `ollaya` command, then the local HTTP API on port 11435. The skill carries rules for when confidence is enough to act and when to hand the question to a person. The MCP server starts with `ollaya mcp`. It provides the `decide`, `list_models`, `show_model` and `pull_model` tools, and question sets appear as `ollaya://presets/<name>` resources. The built-in sets are triage, email, guard, moderation, router and agent. The last one reviews a command the agent is about to run.

Who it is for. For agent developers who want to classify incoming messages, filter requests or vet risky commands quickly and locally, without spending LLM tokens.

Good fit when

  • You need to sort a hundred tickets by team and urgency, and calling an LLM for each is too costly
  • Before running a command the agent must get an allow, ask or block verdict from the `agent` preset
  • You want to filter jailbreaks and prompt injections with the `guard` preset while data stays on the machine
  • You have a client for TypeSafe Jev and want to point it at a local server by setting `TYPESAFE_BASE_URL`

Not a fit when

  • You need new text such as a summary, an explanation or a reply to a customer: decision models generate nothing
  • The answer does not reduce to a known set of options, a rating or yes and no
  • The machine lacks the hardware: large models such as `jeeves` and `clef` need a 24 GB GPU

Example request

Run these 200 tickets through ollaya with the triage preset and show a table: ticket, team, urgency and probability. List the uncertain ones separately

Limitations

  • Ollaya must be installed and models downloaded. The first call pulls the model, and without an Nvidia GPU the developer advises starting with `laya`, which answers in 0.2–0.4 s on a CPU.
  • Model weights belong to their authors and load from Hugging Face. Terms differ: for example, `nimble:9b-v3` ships under a non-commercial license.
  • Accuracy is moderate: the recommended `winnow:e4b` scores 0.724 on typed decisions against 0.738 for the closed Jev, and `von` is archived upstream.
  • Decisions apply only when confidence is high. By the skill, the `choice` threshold is 0.6 and the `noul` thresholds are 0.8 and 0.2; anything in between must be escalated.

How to disable. Remove the MCP server with `claude mcp remove ollaya` and delete the skill from the agent's skills directory. `ollaya stop` stops the daemon, and `ollaya rm <model>` deletes a model.

FAQ

How do I install Ollaya in Claude Code?

Install Ollaya with `curl -fsSL https://ollaya.dev/install.sh | sh`, then run `claude mcp add ollaya -- ollaya mcp`. Install the skill with `npx skills add ollaya-dev/ollaya --skill ollaya-decisions`.

How does a decision model differ from an LLM?

It generates no text. The input is a state and typed questions, the output is probabilities, all in one forward pass within milliseconds. Keep LLMs for tasks that need new text.

How is Ollaya different from Ollama?

Ollaya runs open decision models from different authors, and its CLI follows Ollama's pattern: `run`, `pull`, `list`, `ps`, `rm`. Ollama itself also accepts `/v1/systemone` requests and runs nine decision models in its own GGUF builds.

Which ready-made question sets exist?

Six: triage, email, guard, moderation, router and agent. Each defines its questions and a state field, such as `message` for triage.

Does data leave the computer?

Decisions run locally. Only model weights and small ONNX graphs from the Ollaya registry download from the network. Request texts are not sent to an external server.

Discussion

No questions or reviews yet.

Skills

A terminal MCP client for local Ollama models: MCP tools, prompts and resources, an agent mode and model switching

CLIExtended access836

One agent launches others in the background: Claude Code, Codex, Copilot CLI, Grok Build and Pi, with a model chosen per stage.

SkillBroad access119

Skill, plugin and MCP server: fast Jev and Laya models make choices instead of a pricey chat model in browsers, games and routing.

PluginExtended access182

MCP server that gives an agent probabilities and a pick from set options, like ticket urgency, so code can branch on a number

MCP serverExtended access361

The cascadeflow skill and MCP server: a cheap model answers first, a stronger one steps in on doubt, budget stays in check

SkillExtended access3.9K

README in short

The README presents the project as a decision layer for agents, describes a self-hosted API compatible with TypeSafe Jev and compares it with Jev and Ollama. It continues with features, a table of models, benchmark results, install methods for Linux, macOS, Windows and Docker, and the repository layout.

README badge

Original SKILL.md

---
name: ollaya-decisions
description: "Make typed, calibrated decisions about text or JSON with local decision models served by Ollaya: classify (choice), rate (score) or check a yes/no statement (noul) in milliseconds, with probabilities you can threshold."
license: Apache-2.0
metadata:
  homepage: https://ollaya.dev
---

# Typed decisions with Ollaya

Ollaya runs *decision models* on this machine. A decision model reads a **state** (a message, an email, a ticket, any JSON) and a set of **typed questions**, and returns a calibrated answer to each question in one forward pass. It never generates text.

Reach for it when the answer is one of a known set of outcomes and you will act on it: route, label, block, escalate, pick a template. Keep reasoning in text for open-ended work.

## How to call it

1. **MCP** (tool `decide`, server `ollaya`)
2. **CLI** (`ollaya` on PATH)
3. **HTTP** (`http://127.0.0.1:11435`), TypeSafe-compatible

Report a problem