Reusable components for a company operating system. Start with a product to see what it does and how to use it. Below those, explore the implementation foundations and third-party tools we build with.
When a recurring spreadsheet job needs a tool of its own. Your workspace agent builds the logic and interface; your company owns the code, deployment and data.
Self-hosted · public release in preparation
Explore WorkbooksConversation-first data collection for interviews, intake and feedback. A separate Spark product, not a third-party integration.
Spark product
Visit unformal.ai New tabA structured starting point for your company's knowledge, workflows, decisions and permissions. Give people and agents shared context before connecting the tools.
Workspace foundations · setup by agreement
Explore the workspace starterLibraries, patterns and tools we reuse in delivery. These support the products and implementations above; they are not all standalone products.
A Rust engine for Open Knowledge Format knowledge graphs: lint the graph, query typed relations, export it for a website, render it. Runs the Belgian accounting graph and a client's public knowledge graph.
Rust · MIT
The delivery patterns we reuse across clients: the agent workspace backbone, an agent workspace UI, a public marketing site with an agent layer, an agent-friendly wiki, multi-app feedback, bank integration via Ponto.
Patterns · proven
Command-line tools for the recurring admin of a Belgian business: tasks, CRM, banking, e-invoicing over Peppol. Agents call them the same way people do, with the same guardrails.
Tooling · in use
Every page served as markdown on request, an MCP server, a public API, discovery files. The pattern behind the page you are reading.
Public information · read-only
Explore the agent layerFewer tools, each with real autonomy. These are the external tools that integrate cleanly with an agent workspace and pull their weight next to what we build. Each entry says what it is for and why it is the one we would reach for.
Turns AI-generated HTML into shareable presentation links, so research and analysis land as readable decks instead of text dumps.
Agents publish a deck with a single command, automating generation through distribution. Per-recipient tracking and version history that a wall of text can't offer.
Clients comment directly on a live site; an agent turns those comments into pull requests for developer review.
Built to orchestrate agent-driven development: feedback in, PRs out, with developers keeping full control through review before anything ships. Kills the scattered-feedback-to-code grind.
Peppol-certified e-invoicing: send and receive electronic invoices and procurement documents over the Peppol network, end to end.
A clean REST API an agent can drive directly, open-source, end-to-end encrypted, and cheap (free up to 25 docs/month, volume-based after). Does exactly the one job a Peppol integration needs.
Secrets management: central store for API keys, certificates, and credentials, replacing scattered .env files across every environment and workload.
Identity-based auth lets an agent fetch secrets just-in-time without a human pasting keys, the safe way to give agents real access. API-first, open-source, with rotation and an audit trail.
Bring one process that is slow, fragile or hard to change. We will look at what a useful first version could do.
Scope, cost and operating responsibilities are agreed before a build starts.