Voice agents on your phone lines
Answer inbound calls, qualify the caller, book the slot and place outbound follow-up calls. The transcript, the decision and the record it wrote all land under the same run id.
The AI control plane you can audit.
Arakno runs voice, messaging and back office agents across your systems, and records every action, approval and source.
Arakno answers on the channels your customers already use, and writes into the systems your team already runs.
Telephony and further CRM, ERP and helpdesk connectors are built into your deployment during onboarding. Where a system has no usable API, that connector is quoted separately before work starts.
An assistant answers. A control plane is accountable.
The moment an agent touches a customer, a calendar or a ledger, the question stops being how good the answer was. It becomes who allowed it, what it used, and what changed.
A chatbot gives you
- An answer, with no record of what it read
- Permissions that live inside a prompt
- One model for every task, at one price
- Silence when it invents something
- A screenshot when your auditor asks
Arakno gives you
- An action, with the source and the id that produced it
- Risk classes R0 to R5, enforced outside the prompt
- The right model per skill, with the cost attached
- Contradictions surfaced for a human to resolve
- A ledger that cannot be edited or deleted
Capabilities
Agents that do the work, not just describe it.
Messaging agents with a send gate
WhatsApp, Instagram, Telegram and email, handled by named personas. Simple replies go out on their own. Anything risky is held as a draft until a human releases it.
Writes into your CRM
Leads, notes, stages, appointments and follow-ups written back into the system your team already uses, through connectors built for your stack.
Answers with citations
Company knowledge is ingested, versioned and cited. Every answer carries the page and the snapshot it came from, so an uncited claim is structurally impossible.
Research with a budget
Web, RSS, code and video sources gathered under an explicit query and time budget, scored by how many independent sources actually corroborate a finding.
A team of department agents, coordinated
Switch on an agent per department, from sales and support to finance and quality, and let them hand work to each other. A booking that becomes an order moves from sales to operations with the handoff recorded, and anything that needs a person waits for a named approver. Overnight, a review reads the counts behind the day and proposes at most four evidence backed improvements for you to accept or dismiss.
Every action is classified, approved and kept.
Governance is not a policy document here. It is a service every other part of Arakno has to pass through before it can act.
Run
Event
Decided by
Class
run 4c17ab
Reply drafted for an unpaid invoice question
policy engine
R2
run 4c17ab
Refund offer held, approval requested
gate
R4
run 4c17ab
Approved by the finance lead, sent
named human
R4
run 4c17ab
CRM record updated, artifact stored
policy engine
R3
Example entries. In a live deployment the ledger is append only: updates and deletes are blocked by the database itself, not by application code.
Classified before it runs
- Risk classes R0 to R5Each tool and send path carries a class. An unknown path fails closed to approval rather than proceeding.
- Policy outside the promptA model cannot talk its way past a rule that lives in the policy engine.
Approved by a person
- Verbatim draftsThe approver sees the exact text or payload that will go out, not a summary of it.
- Staged enforcementEnforcement moves from observe, to consult, to fail closed, one reversible step at a time.
Kept for the audit
- One correlation idTask, run, tool call, artifact, approval and audit entry all resolve from a single id.
- Secrets by referenceCredentials are registered and rotated by reference. The governance store has no column that could hold a secret value.
Two ways in, one platform underneath.
A ten person company and a regulated group do not need different products. They need the same spine with a different amount of it switched on.
Growing teams
You are losing calls and messages after hours, and your team retypes the same information into three tools.
- One agent live on your busiest channel first
- Appointments, quotes and follow-ups written back automatically
- Approval on anything that touches money or a promise
- Run by us, so nobody on your side has to operate it. You only bring a model account
Enterprise
You can only deploy AI where access, evidence and human authority survive an audit.
- Single tenant deployment, isolated per client
- Risk classes and approval routes mapped to your delegation rules
- Data rights register, retention windows, export and purge
- Evidence your auditor can read without asking us for a screenshot
How a deployment goes live.
Nothing is switched to autonomous on day one. Arakno earns each permission in front of you.
Scope and connect
We agree the channels, the systems and the actions in scope, then connect them with consent recorded per source and credentials registered by reference.
Run in shadow
The agent handles real traffic while every outgoing action is held as a draft. You read what it would have done before it is allowed to do it.
Release, class by class
Low risk actions are released first. Higher classes stay gated until you decide otherwise. Every step is reversible with one setting.
Questions buyers ask first
Where does our data live, and who can reach it?
Each customer runs on an isolated single tenant deployment. Knowledge, memory and records are scoped to that tenant, internal services are bound to the loopback interface, and the operator surface sits behind mutual TLS and authentication. Credentials are held by reference and rotated on a schedule.
What stops an agent from doing something it should not?
Three things, in order. The policy engine classifies the action before it runs and fails closed when the path is unknown. The approval service holds anything above your threshold as a verbatim draft for a named person. The audit ledger records the decision either way, and cannot be rewritten afterwards.
Can it work with the CRM and phone system we already have?
That is the normal case. Messaging channels, calendars and spreadsheets are connected out of the box, and telephony and CRM, ERP or helpdesk connectors are built into your deployment during onboarding. Where a system has no usable API, that connector is quoted separately before work starts. We map to your existing records rather than asking you to migrate.
How do you handle KVKK and GDPR obligations?
Personal data is detected on ingestion, sources carry a sensitivity level and a retention window, and each data subject request resolves to a per source export or a two phase purge. The processing agreement and the retention terms are agreed before the first channel goes live.
What does it cost?
For growing teams, 120 USD per department agent per month, VAT included, with no setup fee. Model usage runs on your own model account, so your provider bills you directly. Enterprise deployments are quoted annually after a scoping call, because their cost depends on your systems, your volume and the support window you need.
What happens if we want to leave?
Your knowledge base, records and artifacts are exportable per source, and the deployment is yours to shut down. Nothing about the design requires you to keep paying to read your own history.
Bring one real workflow. We will show you the trace.
A scoping call takes forty minutes. You leave with the actions Arakno would take, the ones it would hold for approval, and what the first month looks like.