Kaidera

Technology That Turns
AI Workers Into A Team

A product-level map of the systems that coordinate AI workers, preserve context, route models, apply steering, verify work, and keep humans in control.

Meet Your AI Worker Team

Five specialized AI workers with clear roles, skills, and boundaries. The story starts with the team, then shows how work moves through orchestration, memory, review, and delivery.

K

Keith

Lead -> CPO

  • Product strategy & specs
  • Task decomposition & delegation
  • Cross-worker coordination
P

PROMI

Master Orchestrator

  • Pipeline orchestration
  • Change detection & sync
  • Circuit breaker management
S

Sophi

Backend Engineer

  • Python API framework & relational database
  • Service architecture
  • Database migrations & testing
M

Marv

Frontend Engineer

  • Frontend applications
  • UI components & visual workspaces
  • Responsive design systems
S

Sage

Knowledge Keeper

  • Documentation & guides
  • Architecture decision records
  • Knowledge graph management

PROMI — The Coordination Layer

PROMI watches the project event stream, routes work, manages structured handoffs, schedules follow-up, catches drift, and escalates when human judgment is needed.

🔀
Task Routing
Right task -> right AI worker, with context
🤝
Structured Handoffs
Clear transfer of summary, evidence, and next step
🛡️
Quality Gates
Spec, review, verification, and approval checkpoints
Drift Checks
Detects stalled or off-track work before it cascades
📅
Scheduled Work
Recurring tasks, reminders, and follow-up loops
🏛️
Human Escalation
Important decisions move to review instead of guesswork

Cortex — The Operating Layer

Cortex is where the AI worker team stays aligned: shared memory, live project communication, structured handoffs, behavioural steering, guardrail evidence, and boot context.

Memory

AI workers pick up prior context without asking users to restate it

Communication

Status, decisions, and handoffs move through one project stream

Steering

Rules and roles are available where AI workers actually work

Guardrails

Evidence and approvals are connected to the action they protect

Handoffs

Work transfers with summary, owner, next step, and supporting evidence

Boot Context

Each session starts with the right briefing, not a blank page

Cortex Memory — How Context Compounds

Cortex memory is the second page of the Cortex story: it explains how decisions, lessons, artifacts, handoffs, and communication become usable context for the next session.

L6 Boot Context

Compact briefing that gives each session identity, critical facts, recent history, and next work.

Start informed

L5 Artifact Memory

Screenshots, diagrams, documents, files, and evidence linked back to the work that created them.

Evidence attached

L4 Knowledge Graph

People, systems, decisions, concepts, and relationships connected into a project map.

Relationships

L3 Code Graph

Code relationships, callers, impact, and review context so changes are not made in isolation.

Impact aware

L2 Semantic Recall

Meaning-based retrieval across lessons, decisions, knowledge, messages, and artifacts.

Find by meaning

L1 Verbatim Record

The exact source record: decisions, lessons, knowledge, messages, handoffs, tasks, and artifacts.

Source of truth

Cortex Knowledge — The Project Brain

Once Cortex remembers the work, it connects the work: people, systems, files, requirements, decisions, risks, and evidence become a relationship map AI workers can use.

Capturework signalsConnectrelationshipsVerifysource trailReusenext taskRequirements, decisions, people, systems, artifacts, and risks stay connected
Relationship Map

Cortex links business context so future workers understand what each item affects.

Reusable With Evidence

Knowledge can be reused because it remains attached to source decisions and artifacts.

Three-Layer Behavioral Steering

Organization rules, AI worker identities, and task-level controls shape behaviour before output is generated. Steering makes worker output predictable without asking users to manage prompt internals.

STRUCTURALPROMPTACTIVATIONAI WORKERBEHAVIOR
Structural

Identity files, role constraints, team hierarchy

Prompt

Task context, phase awareness, skill templates

Activation

Runtime triggers, circuit breakers, escalation

Structured Execution, Not Random Prompting

Every task follows the PREVC protocol — five phases with verification gates. AI workers prepare, research, execute, verify, and commit with evidence instead of improvising from a single prompt.

P
Prepare
Gather context & constraints
R
Research
Explore codebase & docs
E
Execute
Implement the solution
V
Verify
Test & validate output
C
Commit
Ship with evidence
Other AI Tools

"Here's some code, good luck"

Kaidera

Structured phases with verification gates

Harness Engineering

The harness gives AI workers a controlled workspace: approved tools, clear permissions, evidence capture, rollback points, and review gates before work moves forward.

1

Scope

What the AI worker can work on and what stays out of bounds

2

Tools

Only approved capabilities are available for the task

3

Evidence

Screenshots, tests, notes, and files stay attached to the work

4

Review

Human approval remains the promotion gate for important changes

Configure Model Capabilities Per AI Worker

Model options come from admin and provider configuration, including managed providers and BYOK. Different AI workers can use different reasoning, coding, vision, voice, and embedding capabilities as the platform evolves.

AI Worker Model Configuration
per worker
K
Keith
Strategy
GPT-5.2
fallback: Claude Opus 4.1
S
Sophi
Build
Claude Sonnet 4
fallback: GPT-5.2
M
Marv
UI
GPT-5 mini
fallback: Claude Sonnet 4
S
Sage
Knowledge
Gemini 2.5 Pro
fallback: Qwen3 class
Cost PriorityQuality Priority
Per-Worker Control

Assign primary + fallback models to each AI worker based on role

Cost Optimization

Use premium models for critical tasks, efficient models for routine work

Intelligent Model Routing

The platform routes work by task type, model capability, provider health, cost profile, and customer configuration across reasoning, code, vision, voice, embeddings, and managed model programs.

Requestfrom workerModelRoutercapability + providerReasoningVisionVoiceResponsebest result
Priority

Preferred model first

Price

Lowest cost per token

Quality

Best capability match

Latency

Fastest response time

Managed Model Programs

For unique or regulated use cases, Kaidera can deliver managed fine-tuning, training data preparation, dedicated endpoints, reserved or on-demand GPU/TPU capacity, geographic controls, and isolated model deployments through approved partners.

Prepare data

Shape training and evaluation sets from approved examples, policies, terminology, and review notes.

Fine-tune

Specialise a model around your documents, workflows, decisions, and quality expectations.

Reserve capacity

Plan dedicated GPU or TPU capacity for predictable throughput, evaluation, and production use.

Burst or isolate

Use on-demand compute for experiments or isolated deployments where enterprise policy requires it.

Vetted Capabilities, Not Blind Plugins

Every skill goes through a five-stage vetting pipeline before it can be used. Parsed, audited, sandboxed, approved by a human, then published to the Kaidera Skills Marketplace.

Parse
Analyze skill manifest
Vet
Security & capability audit
Sandbox
Isolated execution test
Approve
Human sign-off
Publish
Live in Marketplace
Kaidera Skills Marketplace

Every skill is vetted, sandboxed, and version-controlled. Open source on GitHub. No blind plugin installs. No supply chain risk.

Draw — Architecture Whiteboard

A self-hosted whiteboard for architecture sketches, process maps, and shared project diagrams. Built on the brilliantExcalidrawopen-source project, integrated with Kaidera project access and secure vault storage.

platform-architecture.excalidrawVAULTWorkbenchOpen diagramDrawProject VaultSave source fileLive project collaborationYouTeam

Live Testing & Demo Environments

Publish a temporary production-like environment when the work is ready to inspect. Test it, demo it, capture evidence, then promote it or let the review environment expire.

Build Ready
finished version
Review Space
temporary
Test Or Demo
real experience
Decision
promote or destroy

A controlled review environment gives users a working version to test or demonstrate, then it expires when the review window closes.

Enterprise Security, Not A Model Wrapper

Data protection, human approval, audit evidence, cybersecurity review, managed model programs, and enterprise deployment choices are part of the platform design.

Data
protected
Access
governed
Models
controlled
HUMAN
approves
Audit
evidence
MFA + SSOPrivate model optionsAudit evidenceCyber review loop

AI Workers Built for Your Organisation

Beyond the five core AI workers, configure bespoke workers for any domain-specific job. Start from one of four pre-built Business Process Templates (IDP, Requirements Analysis, Predictive Analytics, Dev Request Handler) or build from scratch with 9 configuration knobs. MCP Server Registry connects external tools to any worker.

Examples
  • Requirements gathering specialist
  • Legal contract reviewer
  • Data visualisation consultant
  • Compliance auditor
  • Domain expert in your industry
9 Configuration Knobs
  • Identity — who the worker is
  • Models — which AI capabilities power it
  • Steering — behavioural constraints
  • Tools — what it can call
  • Autonomy — how independently it acts
  • Connectors, Knowledge, Interface, Handoff
How It Works
  • Describe what you need to Keith
  • Keith configures all 9 knobs via API
  • Your worker is live immediately
  • Gets smarter with every session
  • Decommission when no longer needed

The same platform intelligence that powers Keith, Sophi, and Marv is available to any AI worker you configure. Same memory. Same steering. Same human-in-the-loop controls.

Transparent Usage-Based Billing

Pre-paid credits with hard stop at $0. Per-model token metering. Itemized invoices. Auto top-up with safety limits. No surprise bills, ever.

Pricing
  • Starter, Pro, and Enterprise tiers
  • Pricing coming soon
  • Enterprise: contact sales
  • Annual billing available
Metering
  • Per-model, per-project, per-user
  • Real-time credit balance
  • Auto top-up (max 3/day safety)
  • BYO API keys (Pro/Enterprise)
Invoicing
  • Stripe-powered invoicing
  • Per-model line items
  • Tax-compliant (Stripe Tax)
  • PDF download available

AI Workers That Know When to Ask for Help

Human reviewers are first-class routing outcomes in Kaidera. Seamless AI-to-human escalation with SLA enforcement, context preservation, and automatic resume.

Uniform Interface
  • Humans registered as reviewers
  • Same API, same routing
  • Role-based matching (BA, legal)
SLA Enforcement
  • Configurable SLA windows
  • Reminders at 75% elapsed
  • Breach escalation to manager
Suspend-Resume
  • AI pauses cleanly on escalation
  • Full context preserved in Cortex
  • AI resumes from exact point

Code With Your AI Worker Team, Anywhere

PREVIEW

VS Code extension, MCP Server Registry, and an OpenAI-compatible API — use your Kaidera workers directly from your favourite IDE and tools. More integrations coming soon.

VS Code Extension
  • Worker chat panel in sidebar
  • PKCE-secured authentication
  • Project sync and code actions
MCP Server Registry
  • Register external MCP servers
  • Test connectivity and bind tools
  • Any worker can use external tools
OpenAI-Compatible API
  • Drop-in /v1/chat/completions
  • Personal access tokens
  • Works with any OpenAI client

Ready to Meet Your Team?

We're onboarding select B2B teams for early access. Tell us about your company and we'll be in touch.

Register for Early Access →