Products
Built to run your critical systems.
Every CodexSigma product ships self-hostable, API-first, and secure by default. Here is what each one does, how it works, and why it matters.
QW Payroll & ERP
FlagshipAI-powered payroll for the UK
QW Payroll & ERP is CodexSigma’s flagship enterprise product: a complete, multi-tenant payroll and ERP platform built for the United Kingdom. It replaces the fragile spreadsheet-and-shared-drive payroll workflow with a single governed system that computes pay, files statutory returns, and runs your finance operations on your own infrastructure.
Every payslip is calculated from a physics-based analytics engine that models tax bands, National Insurance, pensions, and statutory payments — so payroll becomes a deterministic, auditable process rather than a monthly fire drill. The platform is designed from the ground up for HMRC Real Time Information (RTI) filing, automatically submitting Full Payment Submissions on schedule.
Because the entire system is self-hosted, your payroll data never leaves your server. No third-party SaaS vendor holds your employee records, your salary data, or your bank details. Multi-tenant isolation means accountancies and payroll bureaus can run hundreds of clients on one deployment, each fully sealed from the others.
Key features
- HMRC RTI automatic Full Payment Submission filing
- Physics-based payroll engine covering tax, NI, pensions, and statutory pay
- Multi-tenant architecture for payroll bureaus and accountancies
- Full ERP modules: ledger, invoicing, expenses, and reporting
- Self-hosted deployment with your data on your own servers
- API-first design for integrations with your existing stack
- Audit trail and tamper-proof payroll records
- Role-based access control and SSO-ready auth
Why it matters
- Payroll that used to take days now runs in minutes
- Never miss an RTI deadline — filings are automated
- Your data stays yours: no vendor lock-in, no cloud hostage-taking
- One platform for payroll, finance, and reporting
- Bureau-ready: run many clients in isolated tenancies
Satellite & supply-chain intelligence
Orbita brings space-grade telemetry and predictive logistics together in one platform. It ingests data from ground stations, satellites, and IoT sensors, then fuses it into a single view of your physical operations — from a container ship mid-ocean to a pallet in a distribution centre.
The platform is built for resilience: telemetry pipelines are fault-tolerant by design, so a dropped link on one path never blinds the operation. Orbita’s forecasting engine models arrival times, congestion, and failure risk, letting you act on disruption before it reaches your customers.
Whether you run a national freight network, a cold chain, or a global manufacturing supply line, Orbita gives you the same visibility operators get from mission control.
Key features
- Satellite, ground-station, and IoT telemetry ingestion
- Real-time fleet tracking with geofencing and alerts
- Predictive logistics: ETA, congestion, and failure forecasting
- Fault-tolerant, self-healing data pipelines
- Interactive dashboards for operations and executive teams
- APIs and webhooks for integration into your own systems
Why it matters
- See the whole chain from orbit to doorstep in one screen
- Predict disruptions and re-route before customers feel it
- Resilient pipelines that keep working when links drop
- Operational intelligence previously reserved for mission control
Autonomous AI cybersecurity
Every company is deploying Cursor, Claude Code, and Copilot — and nobody is watching what they do. Resilio is the first security platform built to protect AI agents. When an agent is prompt-injected and starts reading /etc/passwd or exfiltrating code to pastebin, traditional tools see nothing: to them it is just a text editor.
Resilio builds a behavioral DNA fingerprint for every AI agent it finds. It discovers agents by process name, config directories, and API keys; profiles their normal behavior — tools called, files accessed, endpoints hit, hosts contacted, hours of operation; then continuously compares live behavior against that baseline.
When the DNA shifts — new tools, credential access, data exfiltration, beaconing to a command-and-control host — Resilio alerts and can auto-neutralize the agent in real time. A command suite (hunt, shield, scan, agent, policy) spans endpoint sensors, vulnerability scanning, and policy enforcement.
Key features
- AI agent discovery across processes, configs, and API keys
- Behavioral DNA fingerprinting with Z-score anomaly scoring
- Detects prompt injection, exfiltration, credential access, and C2
- Autonomous response: alert and neutralize rogue agents
- Endpoint sensors for process, network, file, and YARA monitoring
- Network and web vulnerability scanning
- Agent permission and scoping policy engine
- STIX2 threat intelligence integration
Why it matters
- Protect the AI agents your team already depends on
- Catch prompt injection and data exfiltration other tools miss
- Under 1% false-positive rate with anomaly scoring
- From agent to agent: watch what agents DO, not what they ARE
Fine-tuned enterprise LLM
Cipher is a privately fine-tuned language model. Built on Mistral-7B-Instruct-v0.2 and adapted with Low-Rank Adaptation (LoRA), it is trained on your domain so it speaks your language: your terminology, your documents, your workflows.
Because Cipher deploys on your own infrastructure, prompts and responses never leave your control. There is no shared SaaS endpoint, no third-party training on your data, no compliance grey area. It is small enough to run on commodity hardware yet strong enough for real production reasoning.
The fine-tuning pipeline is repeatable: retrain Cipher whenever your corpus changes, and ship a new model the same way you ship a new release of any other component.
Key features
- Fine-tuned from Mistral-7B-Instruct-v0.2 with LoRA
- Domain-adapted on your own corpora and terminology
- Runs entirely on-premise on commodity hardware
- Repeatable training and retraining pipeline
- Standard transformers API — drop-in for your stack
- Self-hosted inference with no third-party dependency
Why it matters
- A model that understands your domain, not just general text
- Your data never leaves your infrastructure
- No per-token SaaS fees and no shared-endpoint latency
- Regain control: retrain and redeploy on your schedule
Quantum-mechanic AI engine
Braine is CodexSigma’s research-grade AI engine: a multi-paradigm neural architecture that fuses thirteen mechanisms drawn from quantum mechanics, chaos theory, holographic memory, biological neuroscience, and reinforcement learning into a single differentiable system.
Its transformer backbone is augmented with superposition and collapse, fractal attention over multi-scale geometry, holographic and FFT memory, dendritic activation, metacognitive trees, a chaotic resonance engine, and a quantum-inspired mixture of experts. The result is an architecture built for reasoning-heavy and financial workloads where ordinary transformers fall short.
On the FinanceBench-HRM benchmark, the fusion architecture more than doubles the accuracy of a vanilla transformer of the same size — 62.5% against 25.0% — with the full model still in training. Braine is the engine behind the financial reasoning products in the CodexSigma portfolio.
Key features
- 13 fused mechanisms: quantum, chaos, holography, neuroscience, RL
- SuperpositionCollapse and FractalAttention layers
- Holographic and FFT memory systems
- Dendritic activation and metacognitive tree reasoning
- Quantum-inspired mixture of experts (QuantumMoE)
- FinanceBench-HRM: 53-task financial reasoning benchmark
- Tool-orchestration agent framework built in
Why it matters
- Reasoning capacity beyond vanilla transformers
- Dramatically stronger financial reasoning on benchmarks
- A research engine that keeps improving with each release
- Backbone for the wider CodexSigma product family
Your five-senses AI teammate
The Robotic Coworker is an autonomous AI teammate built around the metaphor of five senses. It sees your codebase and documents with RAG vector search, hears you with voice transcription, touches the world by executing tools over MCP, smells problems with compliance and anomaly scanning, and tastes the quality of its own output before handing it over.
Everything runs through a single coworker surface: one backend API, one web IDE, and all MCP servers started with a single command. Ask questions in natural language, speak a task, or delegate a workflow — the coworker plans, executes, and reports back.
At its core sits EigenMind, a quantum-reasoning component that runs superposition, interference, and collapse over candidate answers and ranks them with an EigenTrust metric — giving you a coworker that reasons, not just regurgitates.
Key features
- SIGHT: RAG vector search over your own knowledge
- HEARING: voice transcription and natural-language tasks
- TOUCH: MCP tool execution — PRs, email, calendar, and more
- SMELL: compliance scanning and anomaly detection
- TASTE: auto-scoring for style, hallucinations, and completeness
- EigenMind: quantum reasoning with EigenTrust ranking
- 46 tools across file, developer, engineering, and data science
Why it matters
- A teammate that listens, acts, and double-checks its own work
- Delegate workflows instead of pasting between tools
- One surface for knowledge, actions, and quality control
- Reasoning that ranks answers rather than echoing prompts