Vendor-neutral by design
Coordinate OpenAI Codex, Anthropic Claude Code, xAI Grok CLI, GitHub Copilot, Cursor, local models, and private CLIs in one workflow. We have no model to sell—so we are never conflicted about which agent you use.
Ozthra AI Relay
Your agents. Your machine. Your subscriptions.
Ozthra AI Relay is the vendor-neutral, local-first coordinator for the AI agents you already use—Claude Code, Codex, Grok CLI, Copilot, Cursor, local models. Compose governed workflows that run on your own machine, with guardrails, approvals, and an audit trail you own. Bring the subscriptions you already pay for. No new model, no usage meter, no lock-in.
Live in early access — early teams lock a founding rate.
Vendor-neutral Local-first execution Bring your own subscription No usage meter
Default coding workflow
Codex implements
Tests run
Claude reviews
Human approval
PR handoff
Build workflows visually—drag steps, set guardrails and approvals, and save reusable templates. No code or JSON required. Coding is the first proven lane; the same engine coordinates any agents you approve.
Ozthra AI Relay coordinates the AI agents you already use on your own machines—so teams keep vendor choice, approvals, and audit under their own control, without a second usage bill.
The always-on agent era
Scout proved the thesis. Ozthra is the vendor-neutral, local version.
Microsoft has announced Scout—slated for August 2026—putting always-on, governed agents on the mainstream agenda, on Microsoft's cloud, inside Microsoft 365, behind premium licensing. Ozthra AI Relay is the vendor-neutral answer available today: it runs where your work already lives, coordinates the agents you already use, and never sends your code or keys to anyone's cloud.
Runs on your machine, not a vendor's cloud
Coordinates any approved agent—no ecosystem lock-in
Uses the AI subscriptions you already pay for
Guardrails, approvals, and audit you own
Shared memory across agents—local, portable, yours (roadmap)
Why Ozthra
A vendor-neutral control layer for the agents you already use
Always-on agents just went mainstream—but most are locked to one vendor's cloud, one ecosystem, and a usage meter. Ozthra is the opposite: a vendor-neutral control layer that runs locally, on the subscriptions you already pay for, with auditability and approvals under your control.
Bring your own subscription
Use the Claude, ChatGPT, Copilot, or Cursor plans you already pay for. Ozthra charges for the control layer—it never resells tokens and never meters your model usage.
Governed by you, locally
Approval gates, audit logs, command policies, repository allow lists, and redacted summaries—enterprise-grade control that runs on your machine, not in a vendor's cloud.
What a run looks like
One workflow, end to end, on your machine
The default template chains approved agents into a governed delivery loop. Each stage is a step you can edit, reorder, or replace—and a human approval gate stands between automation and anything leaving the machine.
Codex implements
Tests run
Claude reviews
Human approval
PR handoff
Codex implements An approved coding CLI works in an isolated git worktree against your repo.
Tests run Your existing test and build commands gate progress before review.
Claude reviews A second model reviews the diff, catching regressions and risky changes.
Human approval A required gate: a person approves before anything leaves the machine.
PR handoff Ozthra AI Relay opens a pull request with a redacted, audit-ready summary.
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 Compose
Compose a workflow
Drag steps onto the visual builder—agents, tests, conditions, approvals. No code or JSON required.
-
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 Approve
Approve before anything ships
A required human gate stands between automation and any commit, push, or PR leaving the machine.
-
4 Audit
Audit every run
Replay the terminal, inspect diffs, and step through checkpoints—a trail you can stand behind.
Works the way your team works
Custom workflows shaped around your stack
Ozthra AI Relay is a schema-driven engine, not a single hardcoded loop. Compose personas, tools, and gates into workflows that mirror your delivery model—built visually so anyone on the team, technical or not, can run and adapt them.
Custom personas
Define implementer, reviewer, and approver roles so each agent has a clear job and boundaries.
Your repos & worktrees
Agents work in isolated git worktrees against repositories you explicitly allow-list.
Your CI & commands
Reuse the test, lint, and build commands your teams already trust to gate every run.
Your secrets & auth
Official CLI authentication and controlled credential boundaries—no scraping, no bypass.
Approved CLIs & APIs
Mix coding CLIs and model APIs from multiple vendors inside one explicit workflow.
Reusable visual workflows
Build, duplicate, and version workflows on a visual canvas so a proven pattern scales across teams.
Automated & auditable
Every run leaves a trail you can stand behind.
Ozthra AI Relay pairs automation with control: gates decide what proceeds, local execution keeps code on your infrastructure, and each run produces a redacted, audit-ready summary your security and platform teams can review.
Required human approval gates
Per-run audit logs
Redacted run summaries
Local-first execution
Controlled credential boundaries
Command policies & repo allow lists
Not a web-chat scraper No usage-limit bypass Official CLI auth
Managed services
Implementation discipline behind the product
Ozthra AI Relay is the product; Ozthra services get you to production. We combine advisory, engineering, integration, security, and ongoing operations so AI adoption reaches production without losing control of risk.
AI adoption roadmap
Prioritize use cases, define ROI, choose vendors, and map adoption steps for secure enterprise rollout.
Systems integration
Connect agents, APIs, data platforms, internal tools, CI/CD, and identity patterns without brittle one-off automation.
Agent workflow delivery
Implement repeatable workflows for software delivery, operations, knowledge work, approvals, and audit-ready outcomes.
Pricing
Pay for control, never for tokens.
Start free with your first workflow. Bring your own agents and subscriptions on every plan—we charge for the control layer (orchestration, guardrails, approvals, and audit), never for your model usage. Cloud execution is a separate, opt-in add-on.
Free — up to 2 seats, local runner, bring your own agents
Pro · $12/user/mo (5-seat min) — unlimited workflows, no token caps, generous fair use
Team · $29/user/mo (5-seat min) — uncapped runs, governance, distributed runners, team memory
Enterprise — SSO/SCIM (rolling out), audit logs, security review, private deployment
On the roadmap
Where Ozthra AI Relay is headed next
Directional — not yet available, and subject to change. These are the threads we are pulling on, in rough order of focus. Nothing here is shipped, and none of it is promised by a date.
- Shared memory across agents
Local, portable, yours: context and decisions that carry across runs and tools.
- Managed cloud sync & team dashboard early seam
An optional hosted control plane for redacted team visibility. The control plane exists; we are productizing the hosted offering.
- SSO (OIDC / SAML) early seam
Enterprise single sign-on. The auth seams are in place; identity-provider wiring is next.
- First-class adapters for more agents
Native support for more agents (e.g. Copilot, Cursor), plus an open adapter SDK to add your own.
- Hosted / cloud runners early seam
Run workflows on managed infrastructure as an option. Today it is local-first with a remote-runner seam.
- Open template gallery & the Agentic SDLC checklist
Shareable workflow templates and an open methodology for building with agents.
- Signed macOS & Windows builds
Notarized desktop distribution for both platforms.
Enterprise trust
SOC 2 Type II and ISO/IEC 27001 trust artifacts available by request.
Ozthra uses security-first delivery patterns: human approvals, local execution options, redaction, auditability, and controlled credential boundaries. Qualified customers can request access to trust materials.