Permission-scoped reasoning
An agent sees exactly what the person asking is allowed to see. No shadow index, no privileged side channel, no second access model to keep in sync.
The agent-native company OS
One graph for the whole company.
OmniManas runs people, delivery, revenue, customers and knowledge on one shared graph, with agents that reason and act inside every person's permissions. Built from the ground up for teams that outgrow their tools faster than those tools can be reconfigured.
Agents cannot run a company on data they cannot see. That is why the graph comes first.
01
People
person · team · approval
02
Delivery
project · sprint · item
03
Revenue
account · deal · forecast
04
Customers
problem · commitment
05
Knowledge
document · decision · rule
Shared spine
One person graph, one event log, one permission model. A customer, a commitment and a deadline mean the same thing in every surface above—because there is only one of each.
Agents
Read and act across the whole graph, scoped to the permissions of the person asking.
Five departments. One set of facts.
Not five products behind one login. One model of the company, with the surfaces each team needs on top of it. No sync jobs, no nightly reconciliation, no second source of truth to argue with.
SurfaceWhat it runsObjects it shares
People
Leave, attendance, comp-off, onboarding, payroll and policy
Person · team · approval · payslip
Delivery
Agile projects, epics, work items, sprints, boards and velocity
Project · epic · item · sprint
Revenue
Pipeline, deals, forecast, clients and demos
Account · deal · stage · forecast
Customers
Problems tracked from first message through to closure
Problem · commitment · resolution
Knowledge
The governed source of what the company knows
Document · decision · rule
Shared spine
One person graph, one event log, one permission model
Every change attributable
Five systems that agree on nothing.
Every tool models the company from its own corner. The expensive work is the translation between them: reconciling names, chasing ownership, re-explaining context to each new system and to every AI bolted on top.
SystemWhat it knowsWhere it stops
HR systems
Know who works here
Stop at what those people are working on
Delivery tools
Know what is being built
Stop at who is free to build it
CRM
Knows what was promised
Stops at whether it was delivered
Support desks
Know what customers reported
Stop at the team that has to fix it
Bolt-on AI
Knows whatever it was pasted
Stops at your permission model
OmniManas
Models the whole company once
Answers and acts across all of it
Built for agents to operate, not just to chat.
AI was not added on top of a records system. The graph, the event log and the permission model were designed so an agent can read the company and act in it without becoming a second, unaccountable source of truth.
Every agent has an owner
Each workflow is registered to a named person accountable for what it produces. Agents sit in the org chart, not beside it.
Work that outlives the session
A resolution open for three weeks. An onboarding spanning a month. A deal cycle across a quarter. Agents hold the thread because the graph and the event log remember it.
One governed tool surface
Every module is reachable through one MCP surface, so agents gain a capability the moment the module ships, not a quarter later.
Event-sourced accountability
Every action an agent takes is an event with an actor, a reason and a reversal path, written to the same log people write to.
Bounded autonomy per workflow
Each workflow carries its own identity, tools, limits and approval policy. Capability expands by policy, never by prompt.
A customer problem, carried to proof.
Because the conversation, the commitment, the delivery plan, the code and production behavior all sit on the same graph, one customer report becomes a single record: investigated, changed, approved and verified without a copied summary between teams.
- 01
Customer signal
Reversals are missing from the settlement export.
The report, affected account and export sample are linked to the weekly reconciliation commitment.
Conversation · account · export
- 02
Investigation
Intent and runtime behavior are reconciled.
Ledger rules, product documentation, the export path and production traces isolate the divergence and blast radius.
Rules · code · traces
- 03
Governed change
A tested patch is ready for accountable review.
The change, regression test, affected customers and reviewer policy travel together in one Change Passport.
Patch · tests · owner
- 04
Release
Deployment health is watched against the original promise.
CI, rollout state and export telemetry remain attached to the same resolution—not a separate incident timeline.
CI · deploy · telemetry
- 05
Verified outcome
Reversal rows are restored and the customer can reconcile.
Production evidence and customer confirmation close the commitment and become outcome-labelled memory.
Behavior · confirmation · memory
Autonomy grows with evidence, not enthusiasm.
Every workflow has its own identity, tools, limits, approval policy and rollback path. High-consequence decisions stay human-owned, and the person who owns the workflow answers for what it does.
01
Observe
Read signals and assemble the record
No external action
02
Recommend
Propose cause, priority and next step
A person decides
03
Prepare
Draft the change, brief or plan for review
Reviewed before it lands
04
Act
Execute approved, bounded workflows
Policy and rollback enforced
The system changes when the company does.
Most business software assumes a stable org chart, a fixed process and an implementation team to encode both. OmniManas assumes the company will change shape, and that the system has to follow without a fork or a migration.
Agile is the native model
Epics, sprints, boards and velocity are how delivery is stored, not a plugin bolted onto a ticket tracker.
Capabilities switch on per company
Turn a module on for one organisation without a fork, a migration or a release train.
One workspace, whitelabelled
Every tenant gets its own domain, brand and policy on shared infrastructure.
No enterprise UX tax
Dense, keyboard-first surfaces with data in the first viewport. No wizard between a person and their work.
One company. One graph. Agents you can hold accountable.
Run people, delivery, revenue, customers and knowledge on one system, and give AI a company it can actually reason about.