Home / Why Tuvel
What makes Tuvel different.
Three ideas run through the whole platform: Multiplayer AI, Earned autonomy and the Control Plane. Each one is available now, in every bank’s private copy.
Being deployed with a Fortune 500 US bank and one of India’s three largest banks.
- 01
Multiplayer AI
People and agents work in the same room, on the same case and the same record. A lead agent hands work to specialists, and your staff approve on cards.
How it works: Multiplayer AI - 02
Earned autonomy
Every agent holds a level and a results score from its own record. It earns wider autonomy level by level, a person decides each step up, and a nightly check drops a level when the work slips.
How it works: Earned autonomy - 03
The Control Plane
One place where every agent’s authority is set, watched and recorded: levels, budgets, guardrails, a kill switch on every agent and every action on the record.
How it works: The Control Plane
01 · Multiplayer AI
People and agents, in the same room.
People and agents work in the same room, on the same case and the same record. A lead agent hands work to specialists, and your staff approve on cards.
The problem banks have
A loan file, a dispute or a KYC refresh is touched by several people. When each person works with a separate assistant, the context lives in someone’s chat and every hand-off starts again.
How Tuvel solves it
Staff and several agents on one case
The credit officer, the risk reviewer, the desk’s lead agent and its specialists, each by name, in one room.
A lead agent and specialists
The lead splits the work, hands parts to specialists and checks what comes back.
Meetings with agents in them
When a question needs several views, people and agents meet. The meeting ends with decisions and owners.
Approval cards
Anything that commits the bank waits on a card for a named person. Only the button approves.
One shared record
Files, decisions and reasons stay with the case, so the next agent picks up what the last one did.
What the bank sees
A room on a lending desk: three people, four agents and an approval card.
Meera, credit officer@Loan File Orchestrator, can we sanction HL-2041 this week?
Loan File OrchestratorThe Cash Flow and Statement Analyser has checked income against statements, and the Policy and Deviation Checker found one deviation. I have called a meeting with Arjun from risk at 3 pm.
Credit Appraisal Note DrafterPicking up from the income check and the deviation note on the record. The appraisal note is drafted.
02 · Earned autonomy
Autonomy an agent earns from its own record.
Every agent holds a level and a results score from its own record. It earns wider autonomy level by level, a person decides each step up, and a nightly check drops a level when the work slips.
The problem banks have
A risk team has to decide how much an AI agent may do on its own. The same freedom on day one as after months of good work is a hard call to sign. Banks want autonomy that grows with evidence, and shrinks as soon as the evidence changes.
How Tuvel solves it
A level for every agent
Five levels, from L1, Observe, to L5, Lead. Every action above an agent’s level waits for a person. Bank agents start at L1 or L2.
A results score from its own record
The share of the agent’s counted results that went well, shown once 20 or more are counted and refreshed every night.
Wider autonomy, level by level
After 30 days of clean work at its level and a results score of at least 60, the agent is nominated for the next level on a card. A person decides.
A level down when the work slips, and any agent can be stopped
Every night the last 14 days are checked: approval cards dismissed or edited, agent-caused errors, bad ratings from several people and outcomes rated poor. Any one past its line drops the agent one level. Any agent can be paused in one click.
Irreversible actions stay with people
At every level, actions that cannot be undone wait for a person.
Two keys for customers
In Tuvel Advisor, the customer chooses a level for the advisor, inside the ceiling the bank sets for their segment.
What the bank sees
An agent’s level and results score, with the record a person reviews before a step up.
86%Results that went well,
from its 212 counted results
- Level
- 2, Assist
- Days at this level
- 41
- Cards decided
- 184
- Cards dismissed
- 3%
- Agent-caused errors, 14 days
- 1%
- Outcomes rated poor, 14 days
- 0
03 · The Control Plane
Where authority is decided and recorded.
One place where every agent’s authority is set, watched and recorded: levels, budgets, guardrails, a kill switch on every agent and every action on the record.
The problem banks have
With dozens of agents across desks, a risk team needs one place to see what each agent is doing, set what it may do, and show afterwards why it acted.
How Tuvel solves it
A live view of every agent
What each agent is doing now, what it just decided and where the next hand-off is queued. Pause, demote or override any agent from the same view.
Levels, budgets and guardrails per agent
Daily and monthly budgets that stop the agent when spent, and guardrails for what always needs a person.
Scoped actions, with their blast radius
Each agent takes only the actions it was granted. Before a workflow change goes live, a blast-radius view shows what it would touch.
Model routing and a kill switch
Each agent runs on the model set for it, through one routing layer. Any agent can be paused in one click, and its next action is refused.
Maker-checker
Two-person approval for changes to connections and rules.
The record, chained and exportable
Every turn and tool call on the record. Calls to bank systems in a hash-chained trail, with a signed export to the bank’s SIEM.
What the bank sees
The live operations view: what every agent is doing, has decided and will hand off next.
- Loan File OrchestratorRouting file HL-2041 to the credit officerWorking
- Cash Flow and Statement AnalyserIncome matches statements; workings attachedDecided
- Credit Appraisal Note DrafterA deviation on a card for MeeraWaiting
- Sanction to Disbursement TrackerNext hand-off: the pre-disbursement checklistQueued
Underneath all three.
What the agents keep from one case to the next, and whose rules they follow.
Learning from your rulings
When your people approve, correct or turn down an agent’s work, the reason is kept, and rulings become playbooks the agents follow next time.
Institutional memory
Decisions, files and reasons stay with the bank, so agents keep the context across cases, desks and people.
Your rules first
Agents apply your bank’s own rulebook from its knowledge base first, with the regulation as the floor, and cite the rule they used.
Questions.
What makes Tuvel different?
Three ideas built into the platform. Multiplayer AI: staff and agents work the same case in one room, on one record. Earned autonomy: every agent holds a level and earns wider autonomy from its own record, level by level, with a person deciding each step up and a nightly check that drops a level when the work slips. The Control Plane: one place where every agent’s authority is set, watched and recorded.
What is Multiplayer AI?
People and AI agents working in the same room, on the same case and the same record. In Tuvel Rooms, staff and several agents work a case together: a lead agent hands work to specialists, meetings include agents, approvals wait on cards for named people, and the shared record lets the next agent pick up what the last one did.
What is earned autonomy?
Autonomy an AI agent earns from its own record. Each Tuvel agent holds a level from 1 to 5. Every night its last 14 days are checked: if people dismiss or edit too many of its approval cards, it causes too many errors, several people rate its work bad or its outcomes are rated poor, it drops one level. After 30 days of clean work at its level, with a results score of 60 or more, it is nominated for the next level on a card, and a person decides.
Who decides when an agent gets more autonomy?
A person. After 30 days at its level, with 25 or more decided approval cards, under 5% dismissed, no level drop and a results score of at least 60, the agent is nominated on a card, and a person decides. Irreversible actions wait for a person at every level.
What is the Tuvel Control Plane?
The place where authority over every agent is decided and recorded: a live view of what each agent is doing and has decided, levels, budgets that stop the agent, guardrails, scoped actions, model routing, a kill switch on every agent, maker-checker, and the record of every turn and tool call, with calls to bank systems in a hash-chained trail and a signed export for the bank’s SIEM.
Are these available today?
Yes. Multiplayer AI, earned autonomy and the Control Plane are available now, in each bank’s private copy of Tuvel.
See all three on your desk.
We open a room, show an agent’s level and results score, and pause an agent live in the control plane.
Or write to demo@tuvel.co.