Ozthra AI Relay · Private beta

The local, vendor-neutral control plane for the AI agents you already use.

Ozthra AI Relay is a schema-driven workflow engine—not one fixed agent loop. Coordinate approved local CLIs, APIs, and future cloud runners while keeping policy, audit logs, environments, and approvals under your control. Bring the subscriptions you already pay for—no token resale, no usage meter.

Starter template

Codex implements
Tests run
Claude reviews
Codex fixes
Manual approval

Build workflows visually—drag steps, set guardrails and approvals, and save reusable templates. No code or JSON required, so technical and non-technical teammates compose them the same way. This loop is just a starting point.

Vendor-neutral Local-first execution Bring your own subscription No usage meter

Ozthra AI Relay is a visual workflow engine, not one fixed agent loop. Compose personas, tools, and gates on a canvas—then run them on your own machines with vendor choice, approvals, and audit kept under your control.

Starter template

From task to guarded PR handoff

The default software-delivery loop helps teams start quickly. Ozthra AI Relay ships this as a template you can duplicate and adapt—not as the only workflow the product supports.

Codex implements Run approved implementation agents in isolated worktrees.
Tests run Execute configured local test commands and capture outcomes.
Claude reviews Route review work to a separate approved agent.
Codex fixes Feed review findings into the fix loop.
Manual approval Require human review before commit, push, PR creation, or PR update.

Build your own

Compose workflows visually—no code, no JSON

The desktop Workflows view is a visual canvas: drag steps into place, set guardrails and approvals, and save reusable templates. Anyone on the team, technical or not, builds the same way while the local runner validates and persists every definition.

Drag-and-drop steps

Compose agent, review, command, condition, approval, git prep, and summary steps on a visual canvas—then duplicate, save, and run.

Set guardrails inline

Add approvals, allow lists, and command policies as you build, so governance is part of the workflow, not an afterthought.

Resume and retry

Continue from a selected step with persisted outputs instead of rerunning completed setup work.

See it in action

The workflow lifecycle, end to end

From a blank canvas to an audited, approved change—here is how a run moves through Ozthra AI Relay.

  1. Composing a workflow on the visual builder canvas

    1 Compose

    Compose a workflow

    Drag steps onto the visual builder—agents, tests, conditions, approvals. No code or JSON required.

  2. A workflow run streaming live agent output

    2 Run

    Run with live agent output

    Kick off the run and watch each agent work in real time, streaming terminal output as it goes.

  3. Human approval gate before commit and push

    3 Approve

    Approve before anything ships

    A required human gate stands between automation and any commit, push, or PR leaving the machine.

  4. Audit view with terminal replay, diffs, and checkpoints

    4 Audit

    Audit every run

    Replay the terminal, inspect diffs, and step through checkpoints—a trail you can stand behind.

Boundaries

Designed for enterprise agent governance, not vendor workaround games

Ozthra AI Relay is not a web-chat scraper, does not collect AI subscription passwords, and is not positioned as a way to bypass vendor billing or usage limits.

Official tool access

Let official CLIs and APIs handle authentication under user and enterprise control.

Local trust boundary

Run local workflows where repositories, SDKs, secrets, databases, and build tools already work.

Explicit approvals

Keep external side effects behind user approval and audit them as first-class workflow events.

Capabilities

A workflow engine for teams that already use multiple agents

Visually built workflows

Compose repeatable agent, command, test, review, approval, retry, and PR feedback loops on a visual canvas.

Worktree isolation

Keep source work isolated by branch and worktree, including named multi-root workspaces.

Local audit history

Capture events, terminal output, diffs, tests, approvals, checkpoints, retries, and PR records.

Runs where code builds

Use the developer laptop or approved local runner before optional cloud execution is introduced.

PR feedback fix loops

Route selected review comments, checks, and PR feedback into controlled follow-up runs.

Org run analytics

Redacted team dashboard with cost (USD), throughput, success rate, and per-agent metrics—plus exports without raw logs, prompts, diffs, or repository paths.

Pricing model

You bring the AI. We bring the control layer.

Ozthra AI Relay never resells model tokens and never meters your usage. You run official CLIs and APIs under your own account and subscription; Ozthra charges for orchestration, guardrails, approvals, and audit.

Bring your own subscription

Use the Claude, ChatGPT, Copilot, or Cursor plans and API keys you already pay for. No token resale, no markup.

No usage meter

Local runs are never metered for revenue. Pricing is a flat plan for the control layer—not a bill that grows with every run.

Local now, cloud optional

Everything runs on your machine today. An optional cloud runner will be a separate, opt-in, usage-based add-on.

FAQ

Common questions about Ozthra AI Relay

Is Ozthra AI Relay a web-chat scraper?

No. Ozthra AI Relay is a local-first orchestration product for coding-agent workflows. It coordinates approved CLIs and APIs under user control—it is not a web-chat scraper and is not positioned as a way to bypass vendor billing or usage limits.

Can I create my own workflows?

Yes. You build workflows visually—drag steps onto a canvas, set guardrails and approvals, and save reusable templates. No code or JSON required, so technical and non-technical teammates compose them the same way. Ozthra AI Relay validates each definition against the schema and persists and runs it locally; JSON is available as an optional advanced export.

How does authentication work?

Ozthra AI Relay does not collect AI subscription passwords. Official agent CLIs and APIs handle authentication under user and enterprise control.

Does Ozthra AI Relay support conditional fix loops?

Yes. Workflows can include condition steps that route to fix loops when tests fail or review findings are present, then continue toward manual approval and guarded PR handoff.

Can workflows target multiple repositories?

Yes. Ozthra AI Relay supports named multi-root workspaces so steps can target approved local repository roots under allow-list policy enforcement.

What happens after a PR gets review comments?

Selected PR comments, reviews, and failing checks can start a follow-up fix run on the existing PR branch. Guarded commit and push actions still require explicit human approval.

Do I bring my own AI subscription?

Yes. Ozthra AI Relay coordinates the official CLIs and APIs you already use—Claude Code, Codex, Grok CLI, Copilot, Cursor, and local models—under your own account, subscription, or API key. Ozthra AI Relay never resells tokens and does not mark up model usage.

How does pricing work?

Ozthra AI Relay is free to start—up to 2 seats with a fair-use daily run allowance. Paid plans charge a flat per-seat price (billed monthly, 5-seat minimum) for the control layer—orchestration, guardrails, approvals, and audit—not for your model usage: tokens are never metered, Team has no run caps, and Pro includes a generous 2,000 runs per user/month fair-use allowance. An optional cloud runner will be priced separately. See the pricing page for current plans.

How is Ozthra AI Relay different from Microsoft Scout?

Microsoft has announced Scout as an always-on agent for Microsoft 365—slated for August 2026—governed in Microsoft's cloud and licensed through premium Microsoft plans. Ozthra AI Relay is vendor-neutral and local-first, and available today: it runs on your own machine, coordinates the agents you already use across vendors, and keeps your code, keys, and audit trail on your hardware.

Will agents share memory across a workflow?

Shared memory across agents is on the near-term roadmap—local and portable, owned by you, and not tied to a single vendor's ecosystem. Today, Ozthra AI Relay passes structured step outputs between steps and persists full run history locally.

Early access

Use Ozthra AI Relay when one agent, one IDE, or one cloud runner is not enough.

Ozthra AI Relay is live in early access on macOS, with Windows in development. Start free and evaluate multi-agent workflows under one governed operating model.