New Neuor for macOS — the free agent harness

The control layer for
AI workflows

Neuor orchestrates models and agents, measures how they reason, and keeps every step accountable. Set the outcome, the tools, the acceptance rule and the budget — get back an accepted result or a clear exception.

Free forever for single-model work · Bring your own model account · No credit card

Bring the models and tools you already use

Platform

Everything a workflow needs to be trusted

Policy and evidence surround the execution path. Memory never grants a permission; a retrieved document never changes the task.

Orchestration across models

Paid policies route, assign roles, compare candidates, critique, repair and escalate — each with explicit breadth and depth caps. Roles follow measured task performance, not a fixed belief about which provider is best.

claudegptgroklocal

Permissions enforced outside the model

Scoped credentials, sandboxed execution, authorization gates on every consequential write.

Reasoning effort, spent deliberately

Start cheap. Invoke another model only after a failed check. Stop when acceptance is met or revisions stop helping.

Durable memory and skills

Session history, project context and preferences retrieved when useful — inspectable, exportable, removable.

  • 1Task history
  • 2Project knowledge
  • 3Reviewed skills

Background work with visible ownership

Schedules, queues, checkpoints and retries. Sleep, shutdown and lost connections show up in task status.

Product

Free harness. Orchestration when it earns its place

Free is a continuing $0 tier, not a trial. Every paid plan includes Free plus coordination across models — and Neuor recommends single-model execution when it's sufficient.

Free $0

Single-model agent harness

  • Native macOS workspace — projects, approvals, deliverables
  • Persistent memory, searchable sessions, reusable skills
  • Files, terminal and code execution, web research, browser
  • MCP and plugin tools with scoped permissions
  • Scheduled and background tasks with budgets
  • Checkpoints with rollback and exportable run records
Orchestration from $19.99/mo

Everything in Free, across models

  • Automatic model selection and role assignment
  • Parallel candidates, critique, repair and synthesis
  • Escalation to a stronger model for narrow issues
  • Budget, deadline and regression stop rules
  • Team roles, shared policies and administration
  • Independent acceptance checks on every candidate
How it works

From intent to accepted result in five states

Every task has a durable identifier and an explicit state. Review, cancellation and failure are outcomes, never surprises.

  1. 01

    Define

    Declare inputs, outputs, sources, model settings, tool permissions, timeouts and acceptance checks before dispatch.

  2. 02

    Admit

    Validate inputs, check permissions, reserve a conservative budget for the work already authorized.

  3. 03

    Execute

    Run approved model and tool calls with recoverable state at every boundary. Interrupted work resumes.

  4. 04

    Validate

    Independent checks decide if a candidate can be returned. Consequential writes pass an authorization gate.

  5. 05

    Close

    Link the accepted artifact to its checks, cost and open issues. Failed runs keep their records too.

A workflow in practice

Evidence travels with the deliverable

An agency's weekly client brief: approved sources in, a reviewable draft out, every claim linked to its evidence, delivery authorized separately.

sourced-brief.flowliveacceptance rule: every claim cited · max 2 revisions
TriggerMonday 07:00Schedule
RetrieveApproved sourcesRead-only · 12 docs
DraftCandidate briefModel of choice · checkpoint
GateAll claims cited?Independent check
AcceptPackage for reviewBrief + claim links + cost
ExceptionReturn with gapsUnsupported claims listed
CheckpointsRetriesHuman approvalIdempotent writesCost reconciliation
Measurement

Judged by accepted outcomes, not finished runs

Quality and the resources it took are reported together, so improvement in one can't hide deterioration in the other.

Cost per accepted task$0.92failed attempts included
Acceptance rateby categorywith uncertainty intervals
Recovery / regressionfixed · brokenseparate denominators
Latencyp50 · p95human wait time split out

Orchestration has to beat the Free harness on the same tasks, tools and acceptance rules. If it doesn't, you shouldn't pay for it.

Pricing

Simple plans. Your own model accounts

You pay providers for inference directly. Neuor's fee covers the harness and orchestration — no markup, no hidden execution charge.

Free

For individuals doing research, writing, planning and file work with one model.

$0forever
Get early access
  • Complete single-model harness
  • Memory, skills, tools, scheduling
  • Exportable run records
  • Basic safety controls
Most popular

Regular

For one person who wants coordination across models on recurring work.

$19.99/ month
Get early access
  • Everything in Free
  • Cross-model orchestration
  • Candidate comparison and repair
  • Budget and stop-rule policies

Team

For agencies and software teams running shared workflows with oversight.

$299/ workspace / month
Talk to us
  • Everything in Regular
  • Shared policies and roles
  • Reviewer and admin controls
  • Priority support

Also: Developer $49/mo · Business $1,499/workspace/mo · Enterprise annual contract · Implementation projects from $3,000. Capacity and support limits are published before launch.

FAQ

Questions, answered

Anything else — write to us and a person replies.

Contact us
Is Free really free?

Yes. Free is a continuing $0 software tier for single-model use, not a timed trial. You still pay your chosen model provider for inference. Core memory, skills, tools, scheduling and safety controls are never paid-only features.

Which models can I use?

Connect your own Anthropic, OpenAI or xAI accounts, or a compatible local-model endpoint. Choose a model per task on Free; paid plans can route across them automatically.

Where does my data live?

Project state and working artifacts stay on your Mac by default, with credentials in protected system storage. Requests to cloud models send what those operations need — Neuor makes those destinations and retention choices visible.

Can a model change its own permissions?

No. Permissions are set by you and enforced by a policy service outside the model. A generated instruction can request an operation; it cannot enlarge what the workflow is allowed to do.

When is orchestration worth paying for?

When it improves accepted outcomes or lowers the cost of reaching required quality on the same tasks — measured against the Free harness with failed attempts and review time included. Neuor shows you that comparison.

Windows and Linux?

macOS ships first. Windows and Linux follow after the initial release; timing is published once the macOS build is in users' hands.

Early access

Make every unit of AI effort accountable

Leave an email and we'll tell you when the macOS build and design-partner pilots open.

No credit card · Cancel anytime · Your data stays on your Mac