RatioCore Book a demo
RatioCore
Book a demo

The platform

One platform to run your business on agents.

Every RatioCore application, from an agent-run store to an agentic SOC, is built on the same core: how work comes in, the steps it follows, the knowledge and systems agents use, the limits they work within and the people who approve. Reinvent the next function on the same foundation, run it where you choose, and keep the freedom to change models and infrastructure.

What it gives your organization

Reinvent faster. Stay in control. Stay free.

Reinvent function by function

Each new application reuses the building blocks and controls you already approved.

Your limits, everywhere

Access, budgets, approvals and records work the same in every function.

Costs you can cap

Every agent has a budget, tracked spending and an off switch.

No vendor lock-in

Your infrastructure, your choice of models, your data in your own database.

Any channel, one set of controls

Same rules in every conversation.

Whether a request arrives in Teams, Telegram, WhatsApp, another approved channel, an API or a system event, the same permissions, approvals and records apply.

WhatsAppTelegramMicrosoft TeamsOther approved channels, apps and APIs

Examples shown in WhatsApp, Telegram and Microsoft Teams. Each channel is enabled by your administrators and follows your policies.

The product in practice

Six building blocks. One set of controls.

Every RatioCore application is assembled from the same parts. Each part has a clear job, and all of them run under the same access, approval and audit controls.

DESIGN

Build the dialog

Guide people through what the agent needs to know. Use structured intake for work that starts from messaging, a business system or an API.

  • Questions that branch on answers
  • Details and evidence captured
  • Starts from approved channels
ORCHESTRATE

Draw the workflow

Set the sequence, branches and approval steps in the visual workflow builder. Use fixed rules for predictable work and agent judgment where context matters.

  • Fixed rules for routing
  • Agent tasks where judgment helps
  • Approval steps inside the flow
KNOWLEDGE

Give agents the right context

Retrieve from enterprise documents and records with vector and lexical search. Add knowledge graph relationships where the workflow calls for them.

  • Vector and lexical retrieval
  • Knowledge graph relationships
  • Evidence attached to each proposal
ACTION

Connect native, API and MCP tools

Define which tools agents can use and which actions require approval. Extend workflows with integrations, APIs and MCP tools.

  • Native, API and MCP tools
  • Approval set per action
  • Channel adapters for messaging
CONTROL

Assign owners and set boundaries

Tenant and role permissions limit access. Sensitive actions route to the people responsible, and authorization is rechecked before execution.

  • Tenant-scoped identity
  • Role-based permissions
  • Named approvers for sensitive actions
OPERATE

Manage agents after launch

Use a shared agent catalog, release controls and operational views. Inspect runs, decisions and external effects as your fleet grows.

  • Shared agent catalog
  • Release controls across teams
  • Run history and recorded effects

Governance in practice

Follow a decision through to its effect.

RatioCore enforces authority when an agent tries to act, not only in a report afterwards. Every protected action passes three checkpoints.

Before execution

Define authority once

Tenant-scoped identity, role permissions and versioned policies establish who can use a tool and when approval is required. Missing authority blocks dispatch.

At execution

Check the action again

Approval is specific to the intended action. The runtime rechecks permissions just before dispatch, then records the action and its outcome.

After execution

Give operators a record

Follow policy decisions, approvals, tool calls and outcomes in context. Reconcile any uncertain external result before retrying it. Use release controls to manage changes across teams.

Flexibility in practice

Build around the work you already do.

Use rules where work is predictable and agents where it needs judgment. Then choose how knowledge, systems and models fit each workflow.

01

Rules and agents together

Keep predictable steps in a visual workflow. Add agents where context or judgment is needed, and keep a person in charge of the decision.

02

Knowledge that fits the question

Bring in knowledge through vector and lexical retrieval, or use graph context when the relationships between records matter.

03

Systems through approved paths

Connect approved systems through channel adapters, APIs and MCP tools. Each connection adds a data path to review before you enable it.

04

Model routes without authority

Different workflows can use different model routes. Models can reason and draft, but they never gain authority to execute actions.

Run it like production

Costs you can cap. Work that survives failure.

Cost control

Know what every agent spends, and stop it when you need to.

  • Daily token budget per agent; the next turn is refused once it is spent
  • Prompt and completion tokens metered per agent per day
  • Per-turn token budgets with announced degradation
  • A kill switch for any agent: disabling it also stops its runtime

Reliability

Built so that a restart or a busy GPU does not lose work.

  • Inputs are admitted once: retrying the same message never runs it twice
  • Answers are stored before they are shown
  • Final answers are delivered at least once, and possible repeats are flagged
  • GPU capacity is shared through a fair queue, so one busy agent cannot starve the others
  • Model servers fail over to a compatible server when one goes down

From one function to the whole company

Reinvent the next function on the same foundation.

Commerce, security, procurement and maintenance need different agents and rules. Each application keeps its own tools, roles and limits, while sharing one platform for approvals, releases and audit records. That is how a company moves from one agent-run function to many.

See the applications

  • Own tools, roles and limits per application
  • Shared approvals and release controls
  • One audit record across the company
  • Applications ready to deploy, and room for your own

Positioning

What RatioCore is, and is not.

RatioCore is

  • An agentic platform for redesigning business functions
  • A governance and execution layer for agents that act
  • A way to build controlled workflows and dialogs
  • A platform you can run on your own infrastructure
  • Open to the models and systems you choose
  • A foundation for optional business applications

RatioCore is not

  • A system of record that only stores what people did
  • A single chatbot or a collection of personas
  • An AI foundation model
  • An unrestricted tool-running assistant
  • A lock-in to one cloud, model or vendor
  • A requirement to buy every vertical application

The Core

Six questions. Answered once.

Every organization that puts agents into production has to answer the same six questions. RatioCore answers them in the Core, so each new agent inherits the answers instead of rebuilding them.

  1. 01

    Who can order the agent to do what?

    Tenants, identities, groups and grants. Every agent gets its own service identity, and every authorization decision is logged.

  2. 02

    What needs human approval?

    Versioned policies and declared tool risk (read, write, destructive) decide. Risky actions wait in an approval queue.

  3. 03

    What happened?

    An audit trail, turn traces and tool events show what each agent did, step by step.

  4. 04

    What did it cost?

    Tokens are metered per agent per day, with daily budgets and quotas.

  5. 05

    How do we recover when something fails?

    Durable admission and delivery, leases and backup checkpoints keep work recoverable.

  6. 06

    Who is responsible?

    Named principals, separation of duties and decision logs keep ownership clear.

For architects · in the product

Controls you can point to.

01

Workflows never half-apply

Every effect, receipt and journal entry commits in one fenced transaction.

02

Approvals that expire

Destructive steps get an approval bound to the specific run, step and input, with an expiry.

03

Immutable versions

Published workflows and dialogs cannot change; each run snapshots the definition and its hash.

04

Deterministic policy

Deny beats approval beats allow, and every policy can be simulated before activation.

05

Zero-token replies

Dialog respond steps return a fixed answer with no model call at all.

06

An instant off switch

Removing a dialog's binding from an agent switches that dialog off immediately.

For architects · design stance

Strict where it matters. Graceful where it helps.

01

Fail closed on authority

Admission, identity, the execution gate, tool bindings and approvals refuse when in doubt.

02

Fail open on enrichment

Memory, grounding and catalog overlays degrade gracefully instead of blocking a turn.

03

Durable before visible

Nothing is shown to a user that has not first been stored.

04

The LLM proposes, the runtime validates, a human publishes

Applied alike to workflows, dialogs and agent profiles.

Proof of value

What a pilot puts to the test.

A good evaluation tests both what the business needs and what the platform promises, in an environment like yours.

  1. Agent lifecycleCreate, configure, disable and redeploy an agentAudit and revision trail
  2. Workflow controlRun a multi-step process with human approvalStep journal, decision and result
  3. Dialog experienceCollect inputs and hand off to an agentScenario replay and completion record
  4. IntegrationRead and write through a constrained adapterAPI outcome and authorization proof
  5. Model choiceRoute to an approved, compatible modelServed-model check and quality test
  6. MemoryRecall permitted information, reject cross-scope retrievalIsolation and provenance tests
  7. Failure handlingInterrupt and retry a turn, replay durable stateConsistent final state, no unsafe duplicate effects
  8. SecurityAttempt unauthorized admin and tool actionsDenial and audit evidence
  9. OperationsBack up, restore and detect a service failureRecovery test and alert record
  10. EconomicsMeasure a representative workloadTokens, infrastructure cost and outcome metrics

FAQ

Platform questions.

Still deciding? Bring your questions to a walkthrough.

What is the difference between a workflow and an agent in RatioCore?

A workflow sets the sequence, branches and approval steps. Fixed rules handle predictable routing. Agents handle steps that need context or judgment, such as gathering evidence or recommending a next step. Policy and role permissions govern what either can run.

How does RatioCore decide whether an action needs approval?

Versioned policies and role permissions define when approval is required. Sensitive actions wait for a named approver. Approval is tied to the specific intended action and rechecked just before dispatch. If the required authority is missing, the tool call is blocked.

Can agents use our own documents and records?

Yes. Agents retrieve from enterprise documents and records with vector and lexical search, and can use knowledge graph relationships where the workflow calls for them. The evidence is attached to the agent's proposal, so reviewers can check it before they decide.

How do we manage agents once they are live?

Use the shared agent catalog, release controls and operational views. Operators can inspect runs, decisions, tool calls and external effects. If an external change has an uncertain outcome, it must be reconciled before anyone retries it.

Is RatioCore a model or a chatbot?

Neither. RatioCore is a platform for running business functions with agents. It sits between your models and your business systems, runs agents, workflows and dialogs, and decides what they are allowed to do. You bring the models you approve, and you can change them later.

See it on your business

Which function would you reinvent first?

Bring a process your team runs today. In a demo we show how agents could run it, where your people stay in charge, and what a focused pilot could prove.

See it on your workflowBook a demo