You're considering software that will sit in the path of every action your agents take. That's a reasonable moment to want to know what we think and where we won't bend — so this page is a list of positions, each one shown beside what it costs us to keep holding it.
Which is how we found the gap. You don't notice it from the outside.
Digital Spectre started as a team shipping agentic systems for other people's businesses. Workflow automation, retrieval, orchestration, integration — the work went well and the agents did what they were built to do.
The uncomfortable part came later, in the handover conversations. Clients would ask a version of the same question — so how do we know what it's doing now? — and the honest answer was that they'd see the output and very little else.
We were handing over software with standing production access and no supervision, and calling it delivery.
Nothing on the market treated an agent the way an organisation treats a person with the same access: baselined, scored, held when something looked wrong, and on the record afterwards. So we built that, and it turned out to be the more important half of the work.
We still build agents. We just no longer ship one that nobody can watch.
A belief with nothing attached to it is marketing. Each of these costs us something real — revenue, speed, or an easier product to build.
If we ever quietly drop one, this page is where you'll be able to tell.
Nothing blocks on day one. An agent runs traced and unrestricted until you've seen enough of its behaviour to write rules that reflect it.
A slower path to the paid tier. We're deliberately telling people to spend a quarter not paying us, and some of them never move past it.
The enforcement layer deploys into your own account. Your traces stream to your own bucket. We hold no production credentials at any point.
Harder operations, slower debugging, and no data moat. The usual argument for a single-tenant SaaS is that it's easier for the vendor. It is.
Not our models, not a third party's, not aggregated, not anonymised. It isn't a setting and it isn't negotiable at any price.
We give up the flywheel every AI company is told it needs. Our product improves because we build it, not because you use it.
On every tier, including free. Security, operations and compliance reading one record is the entire argument — charging per reader would contradict it.
We hand back the easiest pricing lever in software. Per-seat expansion is how most platforms grow revenue without shipping anything.
Ninety days of access after termination, a full export, and a hash chain anyone can verify without us. Agents keep running even when enforcement stops.
We remove our own switching costs. Every renewal has to be won on whether the product is still worth it.
Low volume with catastrophic failure, outcomes nobody can define, teams who haven't been told. We say so in the first fortnight rather than the fourth month.
Signed deals we don't take. It's the most expensive position here and the one we're most often asked to bend.
Spectre reduces the chance of a bad agent action landing. It does not promise to catch every one, and we won't let a sales conversation imply otherwise.
A weaker pitch than competitors willing to say "never". We'd rather lose that deal than be quoted back this sentence after an incident.
The fair question about a company in our critical path is whether it'll still be here. Here's what we can tell you about that.
Each of these would sell. Each would also make the product worse at the thing it's for.
Tell us which. The positions on this page have been argued into shape by customers, and the ones that survived did so because someone pushed hard on them first.