Services

Service as Software is a service delivered by software. Instead of a tool your team operates, you receive the work done. AI agents carry out the steps, and people approve the points where a decision matters.

A copilot helps a professional work faster, and the work stays with them. In Service as Software, the system carries out the service end to end and your team supervises the result. The model works best for work that follows a script and whose result can be checked, often work that is outsourced today. Where the result cannot be checked, a copilot is usually the right choice.

Implementation is most of the work. Sigil Labs maps the process with your team, defines the quality criteria, the approval points and the limits of each agent, and only then puts the system into production. Each agent can reach only the tools it needs.

Control stays with your team, and the system records what was done, by which agent and on what basis. In VoxxFlow, a Sigil Labs product, agents plan, create, review and publish content, and nothing is published without human approval. In Empenho, a proof of concept, nothing written by the AI reaches a client without human approval.

What you receive

• A map of the process and its quality criteria, built with your team
• An agentic system in production that carries out the service end to end
• Human approval points and limits for each agent, adjustable by your team
• An auditable record of every action the agents take
• Ongoing operation and evolution, if you want it

Projects that demonstrate it

• VoxxFlow In production
• Pontifex Internal tool
• Empenho Proof of concept
• Legal triage assistant for Microsoft Teams Proof of concept
• Early Childhood Copilot Proof of concept

Read more about Service as Software →

Custom software is a bespoke system built around one client's process: web apps, installable mobile apps and internal tools. It is designed and built for the way that client actually works, not adapted from a generic product.

The work starts from the process, not from a feature list. Every record has a status, every rule is written down and every action leaves a trace of who did what and when. The same approach produced an equipment management system that follows each item from arrival in stock to write-off.

Sigil Labs also builds what surrounds the application, so the system can run after delivery. Raya, an agentic financial planning platform under construction for the consultants of an investment consulting firm, joins diagnostics, a pension-plan studio, asset-allocation tools and PDF reports in one application, with its own data pipeline and deploy scripts kept in the same repository.

What you receive

• A web app, installable app or internal tool built around your process
• Roles and permissions checked on the server, with an audit log of who did what
• User manual, administrator manual and a go-live runbook
• Containers and a deploy script ready to run in production

Projects that demonstrate it

• Hadur Early development
• 78Worlds Beta
• Empenho Proof of concept
• ART Controles In production
• Raya In deployment
• Incident management for a major motorsport event Ready for delivery

AI and LLM solutions are retrieval systems, agents and assistants that answer from a client's own knowledge and cite the source of each answer, so that a reader can check it.

The model does not answer from memory. The system first retrieves the passages that bear on the question from your own documents, and only then asks a language model to write the answer from them, so each answer points to the passage that supports it. Retrieval runs in two stages: a search by meaning, then a reranking model that keeps the most relevant passages.

Where a number is needed, code does the calculation and the language model only explains it. Early Childhood Copilot, a proof of concept running in production, works this way: it cites the law behind each answer and computes its municipal diagnosis in code from public data. It is built on Libriq, the Sigil Labs platform that turns books and documents in ten formats into a collection that answers with its sources.

What you receive

• A knowledge base built from your documents, indexed automatically
• Answers that cite the source passage, with a link that opens it
• Language, embedding and reranking models that can each be replaced
• An API and an MCP server so other systems can use the collection

Projects that demonstrate it

• VoxxFlow In production
• Libriq In production
• 78Worlds Beta
• Legal triage assistant for Microsoft Teams Proof of concept
• Early Childhood Copilot Proof of concept

Infrastructure and operations cover infrastructure as code, container deploys and the operation of platforms in production: every change is declared and reviewed before it is applied, and every deploy is checked afterward.

Every infrastructure change is written as code, planned, read in full and only then applied. Nothing is changed in the cloud console, because a change made there is missing from the recorded state and the next apply would silently revert it. The state is stored encrypted and locked, and the same code declares the network, the host, a database with no public access, a load balancer with TLS and private storage.

Images are never built on the server. A deploy first confirms that the image for the latest commit has been published, pulls it and recreates the containers, then checks that every service is up, the health endpoint answers and each worker announces its queue. A rollback points the production tag back at the last good image, and the registry keeps the last ten. VoxxFlow runs this way in production.

What you receive

• The whole infrastructure described as code, with a plan read before every change
• Container images built by continuous integration and kept in a registry
• A written deploy procedure with post-deploy checks and a rollback path
• Server access through an authorized session, with no administration port open
• Daily database backups and alarms for downtime, security changes and spending

Projects that demonstrate it

• VoxxFlow In production
• Raya In deployment

Security and audits cover two separate offers: fixed-scope technical audits of live systems, and a published method for governed internal investigations, in which AI works only inside written authorization and limited access.

Governed internal investigations are a method that Sigil Labs has written down, not a finished product. Nothing starts without a signed authorization and a defined scope, and an approval is valid for one specific scope only. The AI gets narrow access, and every call states its case and its reason. It looks for evidence that points both ways, each batch of evidence is hashed, and the work stops by rule when a condition fails, for example when an authorization has expired. The method does not replace legal advice.

A technical audit is a single engagement with a fixed scope on a site that is already live, run in read-only mode. A senior consultant directs AI agents that run the mechanical checks in parallel, and the work ends in an executive summary, technical requirements and a consolidated report. Our technical baseline audit of an institutional website covers security, structure, performance, accessibility, SEO and editorial flows.

What you receive

• An executive summary with the critical findings and their severity
• Technical requirements with quantified thresholds, ready for a selection document
• A consolidated report that groups findings by severity and sets a priority for each fix

Audit project on this site

• Technical baseline audit of an institutional website Delivered