About Oxytal AI Lab

We build AI agents that do the work — and leave the decisions to you.

Oxytal AI Lab is an applied AI studio building agentic systems that take on real operational work — without ever taking humans out of command. Forge, our platform, runs 24 specialised agents across software delivery, security, estate governance and commerce — each one costed, gated and traceable.

Our mission

Make delivery fast and accountable — not one at the expense of the other.

Shipping software — and running a business — is still slow, repetitive and full of hand-offs. We believe most of that work can be done by AI agents, provided the right human decisions stay in the loop. Oxytal builds the agents, the gates and the observability that make that trade real: the speed of autonomous systems, with the control, visibility and accountability that serious teams require.

Why we exist

The bottleneck was never capability. It was trust.

Modern models can already write, review and operate software. What has been missing is a way to let them — safely, visibly, and with someone accountable at every critical step.

The problem

Delivery is a chain of hand-offs — specs to design to code to review to release — each one a place where context is lost and time leaks away.

The temptation

Fully autonomous AI promises to erase all of that. But hand a black box the keys to production and you trade one problem for a scarier one.

Our answer

Agent teams that do the work, human gates that keep control, and a full record of every decision — so speed never costs you oversight.

What we’re building

One pattern, applied everywhere work happens

Every Oxytal suite is the same idea in a new domain: a team of specialised agents, human gates at the decisions that matter, and complete traceability underneath. Four families today, plus any agent you build yourself.

  • Flagship

    Forge — the AI-run SDLC

    Eight specialised agents take a feature from Confluence requirements all the way to production — design, development, code review, build & deploy, testing, defect fix and monitoring — with three human approval gates along the way.

  • Suite

    Security, compliance & estate governance

    DevSecOps scanning across dependencies, secrets, SAST and supply chain; continuous SOC 2 / ISO 27001 / GDPR evidence with an audit-readiness score; and read-only posture across your GitHub orgs, Vercel teams and SharePoint sites.

  • Suite

    Commerce & operations intelligence

    Eight Shopify agents reading live store data — revenue and forecasting, RFM segmentation and churn, conversion and attribution, inventory velocity and stockouts, and fulfilment SLAs.

  • Platform

    One control plane — and your own agents

    A shared foundation under every product: an agent library, a version-controlled Prompt Studio, cost forecasting, role-based access and real-time observability — plus a builder for spinning up custom agents on any model, trigger and output schema.

What we stand for

Six principles behind everything we ship

They are not slogans — each one is enforced in the product, from approval gates to per-run cost accounting.

  • Human-in-the-loop by design

    Agents propose; people decide. Approval gates sit at design, code review and release — nothing ships unsupervised.

  • Traceable by default

    Every run, token, cost and gate decision is recorded. You can always see what an agent produced and why it passed or failed.

  • Composable, not monolithic

    Pick the agents you need, swap prompts in the Studio, freeze outputs for cheap test runs, and wire gates wherever you want control.

  • Accountable economics

    Real token usage and spend on every run, rolled up into role-based ROI you can take to the board.

  • Your stack, your keys

    Runs against your own Confluence, Jira, GitHub, Vercel, Shopify and Microsoft 365 accounts, with your own model API keys — in your environment, not ours.

  • Multi-model by design

    Run any agent on Anthropic Claude, OpenAI GPT or Google Gemini — chosen per agent to balance capability against cost.

Where we’re headed

The plan, in three moves

We started from a single proven loop — software delivery — and have been building outward from it into the wider world of operational work.

  1. Beyond code, already

    Twenty-four agents in production — the full SDLC as one observable pipeline, plus security, compliance, estate governance and commerce intelligence. Every run costed, gated and traceable.

  2. Deeper reach, provable value

    More native integrations, richer forecasting and ROI reporting, and a more powerful custom-agent builder — so delivered value is provable, not assumed.

  3. An agent workforce for every team

    A control plane where any team composes its own agents for its own workflows — autonomous where it’s safe, human-decided where it matters.

Forge, at a glance

Accountability, by the numbers

The product facts that make the mission concrete.

  • 24
    Specialised agents

    Delivery, security, governance & commerce

  • 14
    Native integrations

    Connected with your own credentials

  • 3
    Human approval gates

    Design · code · release

  • 100%
    Human-signed-off releases

    Nothing ships without you

See the idea in motion.

Open Forge and run a pipeline end to end — agents doing the work, gates keeping you in control, every run costed.