The Future

Arrived since the summer

Some of what this page used to list as future work is now simply how Azquo is used.

A hosted MCP server. Azquo is served over Streamable HTTP at https://data.azquo.ai/mcp. In the Claude app it is added as a custom connector and you sign in with an ordinary Azquo login — no key, no tunnel, nothing to install. Other MCP-capable runtimes connect with an API key from the Azquo admin site.

A library to start from. Every session sees a shared, read-only library of working examples — budgeting, management accounts, time and expenses, pivot analysis, property insurance premiums and a US auto quote — and an agent can copy any of them into your own space. Nobody starts from a blank database.

A safety net for agent writes. An agent can rehearse a risky change on a copy before making it for real, every change to a name is kept in its history and a deleted name can be restored, and earlier values are kept along with who changed them and where the change came from.

Ready but not yet routine

Several capabilities already proven in the work but not yet routine, awaiting their moment.

Portable databases as packages. An Azquo database can be packaged with round-trip scripts so a colleague or partner can unpack it on a foreign machine with a single command, rebuilding the on-disk working folder from the database itself. We have proved the pattern internally. The goal is to make it routine: every customer’s database becomes a distributable artefact that travels with its conventions intact.

Self-documenting institutional memory. Every Azquo instance ships with the Azquo_Skills database — several hundred notes on working with the platform. The server now tells every agent that connects to read its starting note before acting, and to write what it learns, decides and hands over back into the database in the same turn, so the next agent picks it up. That loop runs today by convention, and agents follow it. What is still to come is the server checking that it happened, rather than trusting that it did.

External-data anchoring. Agents can fetch external context — weather, regulatory filings, market data, public datasets — and anchor it to the same hierarchies as a business’s internal data. This turns Azquo from a static store into a live cross-reference layer between proprietary data and the world. Demonstrated; not yet a deployment pattern.

Where a partner could help

Concrete asks, where the right collaborator could shorten timelines.

Deeper Excel integration. The Excel task pane now saves changes back and runs a report’s routines with their progress shown as they go. The next step is Custom Functions, MCP-backed cell formulas and workbook-as-database projections — where the spreadsheet stops being a viewer of database content and becomes a working surface for an agent. Office-extensibility experience would land in days, not weeks.

Agent-harness conventions. The first layer is in place: every agent that connects is given the same short set of critical rules by the server, and each rule points to the note that explains it. What a partner could help build is the rest of the harness — the checks, tool surface and deployment patterns that let a customer point any agent at their Azquo instance and have it operate by convention rather than by improvisation.

Get involved

If anything here resonates — especially if you have built agent harnesses, care about data lineage and audit trails, or have an Office-extensibility instinct — we would like to hear from you.