Greenkeel takes AI work from submission to board to production on one platform. Ideas are checked against everything already in flight, and every coding agent gets an owner, a budget and an audit trail, through one gateway inside your own tenancy.
30-minute walkthrough · Runs in your tenancy, nothing leaves it · Now taking three design partners.
Every model, key and change goes through the same gateway. No shared credentials, no invisible access.
Every tool talks to whatever model it likes, with whatever key is to hand, and nobody can say what happened.
AI governance and enablement
AI ideas are arriving faster than existing teams can review them. Greenkeel runs the whole journey in one place: capture every idea, check it against everything already in flight, give the board a complete pack, and carry approved work into a governed build.
What we're seeing
With Greenkeel
Submit an idea
Citizen developers and engineers describe the problem, the data, who will use it and the tools they want, in plain language.
Check against other ideas
Each submission is compared with existing ideas, approved patterns and live use cases. Overlaps are flagged and the submitter is pointed to what they can reuse or join.
Risk-tier and prepare
Tiered against your policy on data, customer impact and regulatory exposure. Missing information is chased, and a draft architecture view is generated from your system catalogue. Low-risk ideas take a fast lane.
Decide
Your AI review board receives a ready-made pack: the idea, its risk tier, related ideas and estimated cost. The decision, conditions and owner are recorded.
Provision
Approval becomes an environment spec, built automatically: a governed workspace with approved models, scoped keys and a budget, all behind the Greenkeel gateway.
Operate and review
Sized before release, then run with live spend, usage and guardrails traced back to the original idea, scheduled reviews and a clean path to retire.
Illustrative data. Idea intake, matching and board packs are being built with our design partners.
Less pressure on existing teams
Architecture, risk and security only see ideas that are new, complete and ready for a decision.
Fewer last-minute changes
Overlaps and risks are found at the start of the journey, when changing course is cheap.
One register for audit
Every idea, decision, owner and pound of spend on one record, ready for your board or a regulator.
From idea to production
A legal operations analyst has an idea for a contract clause checker. Here is every screen between her first sentence and a governed release.
A citizen developer describes the idea in plain language. Overlaps are checked while they type.
Illustrative screens. Idea intake, review and environment build-out are being built with our design partners.
Connected and searchable
Greenkeel connects to SharePoint and Outlook, keeps every document in a safe space, and indexes submissions, decisions and tasks so the next person with an idea can find what has already been done.
Yes. An NDA triage assistant was approved in August using the same clause library, and data protection has already agreed masking rules for supplier contracts. Reuse both before starting from scratch.
Uses the same clause library. Reviewers asked for confidence scores next to each flag.
Board decision · 14 AugMasked samples approved for development. Party names and bank details removed.
IDEA-0377 · closed 2 SepLimitation of liability: cap at 12 months' fees unless approved by the General Counsel.
Indexed 21 OctIllustrative screen. SharePoint and Outlook connectors, the document store and search are being built with our design partners.
Connected to where work happens
SharePoint for documents and Outlook calendars for reviews. Board meetings are found and booked without an email chain.
A safe space for every document
Files attached to an idea are held in a governed store inside your tenancy, with the same access rules as the idea itself.
Search that knows what came before
Every submission, document, decision and task is indexed. New submitters see what has been tried, approved and learned, limited to what they are allowed to see.
Tasks on the record
Actions agreed at review are tracked against the idea, so the next team can find who did what and how it was resolved.
Systems and architecture
Greenkeel keeps a list of the systems in your enterprise and uses it to generate a layered architecture view for every idea. Architecture and engineering teams see exactly what is new, what changes and what it depends on, before the review starts.
The generated view is a starting point, not a finished design. It gives everyone the same picture to discuss, so architects and engineers can spend their time refining it into a detailed design specification.
Illustrative screen. The system catalogue, generated views and diagramming connectors are being built with our design partners.
One list of systems
Keep a maintained catalogue of the systems in your enterprise, with owners, layers and lifecycle. Start from a spreadsheet or your existing inventory.
Diagrams generated, not drawn
Each idea is mapped onto the catalogue to produce a layered ArchiMate view: business, application and technology, with new and changed elements marked.
Open in the tools architects use
Export to diagramming tools such as diagrams.net, Visio and Archi, using the Open Group exchange format where the tool supports it.
A starting point for the design
Teams begin from a shared draft instead of a blank page, then work it up together into the detailed design and specification the build needs.
The right environment, automatically
The hardest part of enabling citizen developers is giving them somewhere safe to work. Greenkeel turns each approval into an environment spec, builds it, and puts the gateway in front of it, so the guardrails the board signed off are the ones that actually run.
Approval
The board's decision, risk tier and conditions.
Environment spec
Generated automatically: region, template, models, data access, budget, guardrails and expiry.
Governed workspace
Built in your tenancy, opened in the tools people already use, and wired to the gateway.
Scoped keys, no shared credentials
Each environment gets its own gateway key, tied to the idea and its owner, and it expires by default.
Safeguards enforced at the gateway
Only approved models. Prompts checked for personal data and secrets. Budgets that stop, not just warn.
Built in minutes, retired cleanly
No tickets and no hand-built servers. When the idea ends, the environment and its access go with it.
Production sizing
Before release, Greenkeel sizes each idea: the effort to build it, what it will cost to run, the load it must carry and who will support it. The board decides with the numbers in front of it, and nothing reaches production without a budget.
About 120 reviewer hours saved a quarter, based on 300 contracts and the Q3 review sample.
Illustrative screen. Sizing gives an estimate to inform the decision, not a commitment. It is being built with our design partners.
Sized from evidence
Estimates draw on the environment spec, the architecture view and real running costs from similar ideas, not a blank spreadsheet.
Cost against value
Build effort and running cost sit next to the expected benefit, so the board can compare ideas on the same terms.
Ready for production
Load, support and budget are agreed before release, and become the limits the gateway enforces once the idea is live.
Estimates that improve
Actual spend and usage are compared with the estimate after launch, so each new sizing is more accurate than the last.
Chat users, engineers and coding agents all go through one gateway. AI Policy and Governance set the rules, and the evidence gets held.
Staff in chat apps, engineers in their IDE and automated agents all reach models through the same gateway, each with its own key type and rules. Choose the endpoints your organisation trusts, in the cloud or on-premise. Provider keys stay behind the gateway, so there is no unapproved route out.
Group your users by role: developers, contractors, platform engineers, security. Each group gets the models, repositories and guardrails its role allows, and a policy changed once applies everywhere that group works.
Agent is editing 6 files in an isolated checkout of billing-core. Network is scoped to the package registry and the model gateway only.
Configurable guardrails inspect each prompt and completion at the gateway. Every change an agent proposes then runs the required checks, and the merge stays blocked until each one passes.
Runs, model calls and gate verdicts land in an append-only ledger, timestamped and queryable. Who can read prompt content is set by role and audited, so an audit becomes an export.
Agent console
Every request is traced from department to person or agent, key and model, priced, and written to a tamper-evident ledger. Open any agent to see its spend, models, keys and blocked requests.
Who is spending what on models, and through which keys. Every figure comes from the tamper-evident request ledger.
Spend
£284.60
1,624 requests
Tokens
12.9m
12.7m in · 173.6k out
Cost / 1m tokens
£22.06
11 blocked
Month to date
£171.40
of £400 budget · forecast £312
Top 5 of 23 principals · select one to drill in
Agent · PR reviewer
Platform · 30d · agt_ci_reviewer
Spend
£18.40
Share of tokens
5.4%
Tokens
696.4k
Requests
200
Avg / request
£0.09
Blocked
3
Tokens over time
Model mix by tokens
Outcomes 200 requests
API keys
ci-reviewer-prod129 req
ci-reviewer-staging55 req
ci-reviewer-dev16 req
Tasks
GK-41286 req
GK-39813 req
No task101 req
Illustrative data. The live console reads every figure from your own request ledger.
Spend by department, person and agent
See who is spending what, on which models and through which keys, over the last day or the last quarter.
Every outcome on the record
Completed, failed and blocked requests, each traced to the agent and the person accountable for it.
Chargeback finance can use
Month-to-date spend against budget, a month-end forecast, and a monthly chargeback CSV by department or person.
Comprehension debt is the distance between what your systems do and what anyone still in the building can explain. Agents widen it faster than people can close it, because the reasoning behind a change now lives in a conversation that ends when the session does.
You still have to read it. You're just not excavating.
Design partnersThree places available
We are working with a small number of regulated organisations in financial services, government, defence and legal to take governed AI from pilot to production. Design partners get founding terms: implementation at cost and six months of managed service free.
Or email [email protected]
Works with the tools your teams already use
Agents and tools
Models
Deploys to