Neuor docs

Getting started

1. Install Neuor for macOS. Early access is rolling out; request access and we will send the build.
2. Connect a model account. Add an API key for any supported provider. Inference is billed by the provider to your own account; Neuor never resells tokens.
3. Create a project and run a task. State the outcome, the sources and tools it may use, the acceptance rule and a budget. Neuor returns an accepted result or a clear exception.

Concepts

Workspace

The unit of ownership: projects, policies, memory, skills and run records. On Free and Starter the workspace is yours; on Team and Business it is shared and administered.

Task

One declared piece of work: inputs, sources, allowed tools, permissions, timeouts, acceptance checks and budget. Every task has a durable identifier and an explicit state.

Policy

How much reasoning effort to spend and how: single-model execution, or orchestration with breadth (candidate attempts) and depth (critique-and-repair rounds), both capped.

Skill

A reusable, versioned procedure a task can invoke. Skills carry their own permissions; a skill cannot widen what a task is allowed to do.

Run record

The complete, exportable account of a run: policy and workflow versions, every provider and tool call, checks, cost and open issues.

Policies and permissions

Permissions are set by the customer and enforced outside the model. A model's generated instruction can never enlarge the permissions granted to its workflow. Memory never grants a permission, and a retrieved document never changes the task.

• Read, write and send permissions are scoped per task and per tool.
• Consequential writes pass an authorization gate; approval to create a draft never implies approval to send it.
• Credentials live in a scoped credential service and never appear in prompts or logs.

Acceptance checks

Each task declares what success is before execution begins. Independent checks decide whether a candidate can be returned. A result includes the checks performed, the evidence used, unresolved issues and actual cost. When acceptance fails, Neuor returns an explicit exception naming the missing evidence rather than spending more unbounded model activity.

Orchestration

Paid policies coordinate several models on one task. Start with one inexpensive attempt; invoke another model only after a failed check; reserve a stronger model for a narrow unresolved issue. A useful critique names a testable defect; repair addresses it and keeps the previous candidate. Stop rules: acceptance met, budget exhausted, deadline reached, or revisions stop improving.

Run records

Every run writes an event record: workspace → task → workflow version → policy version → provider call → tool action → acceptance result → cost. Records are exportable, so a reviewer, an auditor or another tool can answer what was allowed, what happened and why it was accepted.

Plans and billing

Free covers the complete single-model harness. Starter ($19/mo) and Developer ($49/mo) add orchestration for one person. Team ($299/mo) and Business ($1,499/mo) are priced per workspace with shared controls and administration. Model usage is billed by your providers; see Pricing.

Changelog

Release notes will appear here with the first early-access build.