Core product
Kera Operator
An autonomous agent that operates a computer the way a capable colleague does — given an objective in plain language, it works inside your real software until the job is done.
Task
“Reconcile this month’s supplier invoices against our rate card and raise disputes.”
- 1
Browser
Opens the supplier portal and signs in with vaulted credentials
- 2
Vision
Reads the invoice table on screen, no API required
- 3
Reasoning
Flags 3 line items that breach the contracted rate card
- 4
Spreadsheet
Writes the reconciliation into the finance workbook
- 5
Email
Drafts the dispute note and pauses for human approval
Capabilities
Everything a person does with a screen and a keyboard.
Screen-native perception
Reads any interface as a human would: pixels, layout, state and change over time. Works on web, desktop, virtual desktops and legacy terminals.
Long-horizon planning
Decomposes an objective into hundreds of verified steps, re-planning when the interface or the data does not match expectations.
Reliable action
Precise mouse and keyboard control with element-level grounding, retry policies and deterministic replay of any prior session.
Tool and file fluency
Uses browsers, spreadsheets, email clients, terminals, file systems and internal tools in the same session.
Self-verification
Checks its own work against the objective before completing, and escalates to a human when confidence drops.
Memory across runs
Remembers the layout of your systems, your naming conventions and prior decisions so the second run is faster than the first.
Agent runtime
A runtime built for work that must not go wrong.
Operator is more than a model. The runtime handles isolation, state, credentials, recovery and observability so that agents behave predictably at scale.
Isolated workspaces
Each run gets a clean virtual machine, destroyed on completion.
Credential vault
Secrets are injected at the keystroke layer and never exposed to the model context.
Policy engine
Allow-lists for applications, domains, files and irreversible actions.
Observability
Traces, screenshots, timings and cost per run, streamed to your stack.
Integrations
Works with what you already run.
Operator does not need an API to be useful — but it will happily use one when it exists.
Security & control
Autonomy with a hand on the brake.
Define exactly what the agent may do, what requires approval, and what it may never touch. Everything it does is recorded.
Approval gates
Route any action class to a human reviewer in Slack, Teams or the Kera console.
Immutable audit log
Signed, timestamped record of every action, with frame-level replay for disputes.
Data residency
Choose your region, or run entirely inside your own infrastructure. No training on your data.
Deployment
Three ways to run Operator.
Kera Cloud
Managed agent fleet with autoscaling virtual workspaces. Fastest path to production.
Your VPC
Runs inside your cloud account. Your network policy, your keys, your logs.
On-premise
Fully air-gapped deployment for regulated environments and sensitive data.
Pricing
Priced by outcome, not by seat.
Pilots are fixed-scope and fixed-price. Production is metered on completed tasks, with volume commitments available.
Pilot
Fixed price
- One workflow, scoped in a week
- Production-grade evaluation harness
- Success criteria agreed up front
- 4–6 weeks
Scale
Metered
- Multiple workflows and agent fleet
- Cloud, VPC or on-premise
- SSO, RBAC and audit export
- Dedicated solutions engineer
Enterprise
Custom
- Air-gapped deployment
- Custom model training and fine-tuning
- 24/7 support with SLA
- Joint roadmap
See Operator run your workflow.
Bring one process. We will run it live and show you the trace, the failure modes and the honest limits.