Kaidera An AI platform I designed and built.

Kaidera coordinates specialist AI workers with project knowledge and human review. I built it because the organisations I worked with needed to prove what their AI had done and who approved it.

Visit kaidera.ai What it includes

Why I built it

Regulated organisations have to show their working.

On the EUROCONTROL account and in other regulated environments, every change had to answer the same question from auditors and safety teams: what happened, and who approved it? Most AI tools cannot answer that.

Kaidera builds the answer into how the work runs. Risk classification, approval gates and the audit record are part of the platform from the start. It is also where I test the methods I recommend to clients.

What it includes

The platform, the appliance and the open source terminal.

Kaidera platform

Teams of AI workers that carry out work inside an organisation's own systems. Each task is classified by risk, high impact actions wait for a named person to approve them, and every step is recorded.

Kaidera OS

The same platform packaged to run on the customer's own hardware, for organisations whose data cannot leave their estate.

OpenKai

The open source terminal the operator works in. It runs agent work in both the platform and Kaidera OS, and holds the settings for AI providers, models and credentials.

How it works

Six things the platform does.

Visible reasoning

Every decision shows what the AI worker looked at and how confident it was, so a reviewer can check the working.

Approval gates

High impact actions stop and wait for a named person. The platform enforces the gate, so it cannot be skipped under pressure.

Risk classification

Work is classified by impact before it runs. Low risk work goes ahead and high risk work queues for approval, using the same rule every time.

Exportable records

A full record of who did what, when and on whose approval, ready for internal audit.

Model agnostic

The governance, workflows and records sit above the AI model, so the model can be changed without rebuilding them.

Flexible deployment

Cloud, the customer's own Kubernetes cluster, or Kaidera OS on their own hardware.

Open source

Parts of Kaidera are open source.

The memory system, the skills catalogue and the agent terminal are published on GitHub so anyone can read, use and improve them.

Cortex

MIT

A Postgres-backed memory and coordination layer for AI agent teams: it preserves decisions, lessons and progress across sessions, with project-scoped worker identities, durable handoffs and semantic and graph search. Grew out of Kaidera OS; currently an early release.

View on GitHub

OpenKai

MIT

An open source, terminal-first coding-agent harness and TUI, built on Pi. Choose models, supervise worker agents, or run an explicit Architect, Builder and Judge Fusion to compare difficult decisions across models. Optional Cortex integration adds shared, project-scoped memory.

View on GitHub

Kaidera Skills

Open source

A catalogue of reusable skills for AI agents. Each skill is a SKILL.md file with a security manifest, checked by automated schema and security-manifest validation before publishing; every skill is marked unvetted (trust_tier: unvetted) for the agent or team using it to review. Usable in any harness that reads SKILL.md.

View on GitHub

Want AI with this kind of oversight in your organisation?

AI development services Talk to me