Agents that compound.
Ensemble improves the agents that run business operations. Memory persists, harnesses are versioned, and the ones that evaluate well are the ones you keep. The inference underneath is yours.
The new business OS.
1
Harnesses, versioned
A harness is the agent's instructions, versioned like code. The dashboard ranks the ones that spent, so you can see which version is doing the work.
2
Actors with memory
An actor keeps memory across the work. Open the actor for its memories and its usage. The next unit of work does not start from nothing.
3
Extract, then evaluate
Extractions and evaluations are first-class usage, next to inference on the donut. Strategies show which extraction path spent. Evaluators record whether the result held.
4
See the whole run
Every run leaves a record you can open: spend, extractions, evaluations, and the logs behind them. Governance sits on that record.
How an agent gets better.
Build the harness
The harness is the agent's instructions, versioned like code. Skills, knowledge, and tools sit on that version.
Find the improvement
Extractions pull what the run produced. Evaluators score it and point at the prompt, skill, or outcome to change.
Promote, then iterate
Make the version that holds live. The next change is another version, so the agent keeps compounding.
On demand vs self hosted.
Run Ensemble as the hosted workspace, or host it in your own environment.
On demand
The hosted workspace
£75/mo
Starts at £75 for a shared 500 and 5,000 page imports a month
The 500 is shared across queries, extractions, and evaluations
Page imports are where sources and extractions go into indexes
Slider from 500 to 10,000 a month
Agents, memory, and versioned harnesses
UK data residency and ISO 27001
Cybersecurity & data governance.
Data assessment.
Where does your data live, where does it flow, and what happens when AI touches it? We map every interaction between your systems and AI services to understand the real exposure before anything else.
Which models see your data, what gets stored, what gets logged, what leaves the building. You can't govern what you can't see — so we make every data path visible before a single workflow goes live.
Smart scanning.
AI-generated code introduces new risks — secrets leaked into outputs, OWASP vulnerabilities woven into suggestions, production data surfaced where it shouldn't be. Traditional scanners weren't built for this.
We run AI-aware scanning across your codebase and outputs, catching what gets introduced by the same tools that are supposed to be helping. Continuous, not one-off — every output checked and reviewed.
Governance by design.
Human-in-the-loop isn't a feature — it's an architecture decision. We design new systems with approval workflows, escalation paths, and audit trails built into every layer from day one.
Every agent action traceable, every decision logged, every critical path routed through a human checkpoint. When something needs reviewing, the system stops and asks — it doesn't assume.
UK data residency.
Inference routed through UK and EU regions. Your data persisted in UK data centres, with processing across nearby European availability zones. An evolving landscape as regional coverage expands.
We architect around the current constraints so compliance isn't an afterthought. For regulated industries, public sector, or any organisation where data sovereignty isn't optional, this is the foundation.
FAQs
Start the journey
Ready to ignite a spark? Everything we build starts here — a focused discovery engagement that maps your landscape and delivers a clear plan forward. Whether you're launching something new, reigniting an existing platform, or rewiring how your team works with AI.