Controlled deployment
Each product runs within defined boundaries, from customer-controlled desktops to managed cloud services.
Request a briefingTrust-first applied AI
AI systems for work that has to be right.
Alchemy Loop builds focused AI products for operational data, documents and software delivery - designed around controlled deployment, inspectable evidence and human accountability.
The Alchemy Loop standard
Trust is part of the architecture, not a promise added later.
Each product applies the same operating discipline to a different high-consequence workflow.
Read-only by design
Database products are built to enquire, not alter systems of record.
Evidence attached
Answers and decisions retain the query, page or exact change behind them.
Human authority
AI accelerates the work; accountable people keep the final decision.
The portfolio
Purpose-built products. One control philosophy.
Independent AI second opinions
2nd Engineer
Give AI-assisted development teams an independent second opinion from a different AI, then record what the review changed.
- Reversible Claude and OpenAI roles
- Advisory, read-only review
- Manager dashboard with measured impact
Governed answering for operational data
AI LLMQuery
Produce precise, auditable answers through a multi-layer AI and retrieval architecture - not an LLM wrapped around a database.
- Read-only by construction
- Domain-learned and auditable
- One answer across multiple databases
Documents become structured data
AI DocQuery
Extract traceable values from recorded documents and difficult scans, with every accepted value tied back to its page.
- OCR confidence by page and region
- Quoted evidence for accepted values
- API, batch and SQL-ready outputs
Standalone AI OCR workbench
Redaction Studio
Prepare, inspect and redact document collections in a controlled desktop workflow.
- Customer-controlled desktop operation
- High-clarity, mixed-page-size viewing
- Human approval before finalisation
Managed AI redaction SaaS
Redaction Cloud
Use the Redaction Studio workflow as a secure online subscription service in a protected commercial cloud environment.
- Managed SaaS rather than software purchase
- SOC 2 Type II and HIPAA-compliant hosting
- Customer-specific governance at onboarding
LLMQuery · Governed answering
More than an LLM wrapped around a database.
LLMQuery is a multi-layer AI and retrieval architecture. It combines advanced retrieval, query decomposition, contextual reasoning and data-layer techniques to produce precise, reliable answers that approach research-grade question-answering systems.
Layered controls keep generated work inside a governed enquiry path and protect systems of record.
Schema context, reviewed definitions and business rules help the system answer the question the organization actually means.
Every answer retains its SQL, assumptions and source coverage so reviewers can inspect how it was produced.
LLMQuery Connect can coordinate read-only queries across independent current and legacy systems while preserving source visibility.
2nd Engineer dashboard
See second opinions in flight - and what they changed.
A manager-facing view of activity, timing, usage and graded outcomes. The public simulation uses fictional sample data; real questions, answers and code remain inside the customer environment.
SIMULATION · SAMPLE DATAIndustry solutions
One control model, applied to the work each team already owns.
LLMQuery can be configured around the vocabulary, rules and review expectations of a specific operating environment.
Default operations
Ask operational and portfolio questions across the case-management database without waiting for a custom report.
Accounting and audit
Explore practice performance and permitted client ledgers with read-only controls and calculations a reviewer can reproduce.
M&A diligence
Ask follow-up questions against a target system while preserving read-only boundaries, query evidence and a local activity record.
LLMQuery · governed multi-database answering
One question across systems that still have to coexist.
LLMQuery Connect extends the governed answering architecture across current and legacy databases - one domain-aware question, one auditable answer, and no need to turn Alchemy Loop into another system of record.
- One canonical business vocabulary across sources
- Identity resolved before counts or totals are combined
- Source contribution and coverage gaps remain visible
- Unsafe consolidated answers are refused rather than guessed
Current system
Legacy system A
Legacy system B
LLMQUERY CONNECTCanonical meaning
Identity and coverage
Safe execution plan
Identity and coverage
Safe execution plan
One answer
source breakdown retained
source breakdown retained
Inspectable by design
Do not take the trust story on faith. Check the evidence.
Request tracing makes transmitted context reviewable instead of leaving privacy as an invisible promise.
Database answers retain the SQL and assumptions; document values retain their page; code reviews retain the exact change.
Approval, rejection and escalation remain explicit human acts rather than silent automated repair.
Private design-partner conversations