[ AI ]
AI that runs your agency, on your terms
ROIkeep already has a governed way in for AI agents: they get only the access you grant, and they can never read your secrets. On our roadmap is the sovereign layer that completes it: your own agent, running the whole system in plain language from Slack or Discord, on open-weight models on our server or yours, with nothing sent to an outside AI provider.
[ Available today ]
A governed gate for agents
Every module is reachable through MCP, the open standard for connecting AI agents to software. Nothing gets in without a key you issue, and the key carries only the access you grant.
Access you choose
Each key carries only the scopes you pick, such as your profile, attendance, people, devices, or audit events.
One workspace, 30 days
A key is pinned to one workspace and expires after 30 days. Owners and admins can revoke any key.
Checked on every request
Every request is checked again for membership, role, and access, so a removed person's agent stops working too.
Approval for clocking
Clocking in or out from an agent creates a proposal that waits for your approval in the browser, with a fresh location check.
What it can never do
No agent can read vault secrets or document contents, invite people, grant roles, or change billing.
Your model, your call
An agent's model sees what the agent reads. Grant access only to agents whose model you trust, or run the model yourself.
[ On the roadmap ] [ Run it from chat ]
Run the whole agency from Slack or Discord
Ask in plain language. The system finds the answer or prepares the change, and asks before it acts.
Slack or Discord
Discord for teams that already live there, Slack for teams that pay for it. You pick.
The interface stays
Everything still works in ROIkeep itself. Chat is a second way in, not a replacement.
[ On the roadmap ] [ Sovereign by design ]
Your models, on your server or ours
Every model is open-weight and self-hosted. No outside AI provider sees your prompts or your data, and no provider sends you usage bills.
On your own hardware
Run the models on hardware you own, and the data never leaves your building.
Or on a server we manage
A dedicated server we run for you, the way we run a Private instance today.
Small to large models
Small open models run the orchestration. Larger agencies can choose larger open models.
[ On the roadmap ] [ Controls ]
Nothing runs unchecked
The layer follows the same rules the system already enforces, and adds its own record of every job.
Role checks first
A request is checked against the person's role before any model sees it.
Approval before action
Every change waits for a person to approve it.
A ledger of every job
Every request and action is recorded, with who asked and who approved.
Separate keys
Each part runs with its own credentials, and no model can grant itself permissions.
[ Questions ]
About AI in ROIkeep
Is the Slack and Discord layer available today?
No. It is on our roadmap and we do not quote a date. The governed gate for agents is available today.
Can any AI agent connect today?
Any agent that supports MCP, once you issue it a key with the scopes you choose. Its model sees whatever it reads, which is why the sovereign layer on our roadmap keeps the model on your server or ours.
Which models will the sovereign layer use?
Open-weight models only, run on a server we manage or on your own hardware. We name the models when it ships.
What will the sovereign layer cost?
The price is not set yet. Ask us on a call where it stands.