IntelligenceIntelligence

Your AI is busy. Know what it's up to.

Intelligence intelligence for businesses building with AI and going round in circles.

Part of Solarise.dev

Get your data talking again.

A single unified interface that connects your CMS, CRM and more.

Many businesses use agents to do real work, but a lot of them couldn't tell you what work was done yesterday.

Everyone wants work done more quickly, but no one likes losing track of what's been done.

Accountability and responsibility are still extremely important, even more so now that we're linking agent intelligence into everything.

Businesses run on questions and answers. With AI in the loop, it's becoming harder for the people who matter to make important decisions with facts and figures at their fingertips. We start by planning out the systems that let you see what's going on.

40%+

of agentic AI projects will be cancelled by the end of 2027, says Gartner. Too expensive and no clear ROI are two of its reasons. A lack of thorough risk management is the third.1 Gartner, June 2025

How we'd do it.

The single source of truth.

The big CMSs and frameworks, WordPress and Laravel among them, now come with MCP servers.2,3 We're not just building for humans now, but agents too.

Gating access is even more important for agents than for humans. Exposing data still comes with extreme risks.

We run through your data requirements together and work out what rules and authorisation steps will be needed for your new workflows.

An isometric drawing of four lanes from an agent to your CMS. posts.read passes an open gate. posts.update and orders.refund each pass a gate with a person beside it. users.delete is stopped at a closed gate.
1 in 4

UK workers spend seven or more hours a week copying information between apps and feeding it to AI tools.4Workday and The Harris Poll, May 2026

Ask for permission, not forgiveness.

On 25 April 2026 a coding agent deleted a startup's production database and its backups in nine seconds. The token it used had been made to manage domain names. Nothing stopped it deleting everything else.5

We put a person in the loop wherever an operation can't be undone. For everything else, we scope each kind of job and give it the credentials it needs.

An isometric drawing: an agent in staging uses a token made for domain names, calls the platform API, and the red volumeDelete call reaches a volume holding both the production database and its backups.
OWASP's 2026 list for agent security calls this Least Agency.6

Work from the same pages.

When people ask for a model trained on their own data, they nearly always want retrieval. It refers back to the source of each answer, and it stays current.

Things fall apart when the index pulls in draft and private pages. The public can end up reading content that skipped the usual checks.

We keep your CMS permissions lined up all the way from the database to the public front end.

66%

of developers say their biggest frustration with AI is answers that are almost right. Retrieval that cites its page lets someone check.7Stack Overflow Developer Survey, 2025

An isometric drawing: published pages pass through a permission gate into the index; a draft and a private page are stopped at the gate. The index answers a person, naming its source page.

An audit log that's useful.

We build a log you can trust, with every action the agent took and whose name is on each approval.

It's written for the person making the decisions, who then has to explain them to someone else.

An isometric drawing of a log as a row of sheets. One entry at 09:15 is approved by the editor; one at 09:22, a refund, is refused as over the limit.

Most businesses don't have a data lake. They have a pond, and there's plenty living in it.

When someone senior asks a question about it, most staff don't have the tools to find the data and shape an answer quickly.

An ink and watercolour sketch of a small pond with reeds, and a duck standing beside it.

Built for you from open-source parts.

Every system we build uses open-source components, tried and tested and maintained by the community. Your foundations will be provably strong and kept up to date.

  1. The connector

    Built on the official MCP SDKs, which cover most languages your system might be written in. It's the standard way to hand an agent a set of operations. We agree the list with you and write each one. If it isn't on the list, the agent can't reach it.

  2. The control panel

    Approvals and settings live in the screens your team already logs into, with roles and a record of every change. If your system has no admin, we add an open-source one.

  3. The search

    Vector search with pgvector, inside a Postgres database, so retrieval sits next to your data. Your CMS permissions travel with every query, so drafts and private pages stay out of the answers. If your data lives somewhere else, it runs as a small separate database.

What do you need right now?

Rough prices.

We'll sit down and work out the cost together. Here's a rough idea of what to expect.

A look at what you've built
£450
A written note on what to fix first, yours to act on with anyone.
Connectors, approvals and a log
£5,000 to £15,000
One system at the low end. Several, with search over your content, at the top.
Finishing something you built with AI
from £2,500
Quoted once we've read the code. Getting a finished site live on your own domain costs less.

15 years of people and systems.

Most past work has been joining systems up and deciding who's allowed to do what. Nobody called it agents at the time.

  • A sync that keeps the Kurt Weill Foundation's website in step with Salesforce. Editors change a record once, and a screen shows them what happened to it.
  • North Commerce's Zapier integration, a safe way for add-ons to reach its core, and the scheduled queue behind its emails, SMS and webhooks.
  • A council's waste-permit system, where two parts held different versions of the same record. We merged them into one audit trail, with the tests written first.
  • The same council's multi-step application service. Clearer routes through the form, tests around each one, and emails that reach the department handling each application.
  • An enterprise expense platform with multi-step approvals. Spending rules could change without touching the approval chain.

Swap the person for an agent and the job is much the same.

Talk it through.

Describe what your agents do now and what you'd like them to reach. You'll come away knowing what to connect first, and you can take that to anyone.

Sources

  1. Gartner predicts over 40% of agentic AI projects will be canceled by end of 2027. Gartner, 25 June 2025. ↑
  2. From abilities to AI agents: introducing the WordPress MCP Adapter. WordPress Developer Blog, February 2026. ↑
  3. Laravel MCP: build with the universal AI standard. Laravel. ↑
  4. UK employees spend nearly a full work day each week managing disconnected AI tools. Workday with The Harris Poll, 14 May 2026. ↑
  5. Cursor-Opus agent snuffs out startup's production database. The Register, 27 April 2026, reporting the founder's own account. ↑
  6. OWASP Top 10 for Agentic Applications for 2026. OWASP Gen AI Security Project, December 2025. ↑
  7. Developer Survey 2025: AI. Stack Overflow. ↑
  8. 2026 GenAI Code Security Report. Veracode. ↑

Talk it through.

Describe roughly what your agents do now and what you'd like them to reach. You'll get a reply with a time, usually within a working day.

This goes straight to my inbox. Nothing is stored here.