Enterprise AI customer support with operational control
Enterprise AI customer support with owner-scoped data isolation, role permissions, append-only audit traces, ownership gates, multilingual multi-step inquiry resolution, REST API, signed webhooks and native MCP server. Metered at $0.006/message.
In short: OctaDezx is an enterprise AI customer service platform that answers customers on chat, messaging apps, email and phone from your approved knowledge, splits compound questions into verified steps, checks that an order or booking belongs to the person asking before discussing it, and records every step for audit. Data is isolated per business in the database, access follows owner, admin and agent roles, and integration runs through a REST API, signed webhooks and an MCP server. It does not claim SAML SSO, SCIM, SOC 2 or ISO certification, or a public uptime percentage.
Bring customer conversations, business knowledge and human decisions into one governed workflow. OctaDezx gives enterprise support teams shared context, controlled access, multilingual coverage across Europe and the Americas and a practical developer path, with the limits of each control stated rather than implied.
The everyday problem
A larger organization needs more than an assistant that sounds helpful. Different teams own policies, customer channels and operational decisions. A wrong promise, an order shown to the wrong customer or a lost handoff matters more at volume than a fast answer, because one mistake repeats across thousands of conversations.
How OctaDezx helps
Evaluate the controls that exist today: business-scoped records enforced by row level security, role and section permissions, approved knowledge, ownership and reference gates, deterministic money and time math, a final verifier, an append-only trace of every step, and human takeover. Pilot representative requests before relying on the assistant for high-impact customer work.
Controls to evaluate in an enterprise pilot
- Data isolation in the database. Each business's records are separated by owner-scoped row level security policies in Postgres, not by a filter in the user interface. Review the policies and your own staff accounts as part of the access assessment.
- Team access by responsibility. Owner, admin and agent roles govern dashboard access, with per-member section permissions. Approving what the assistant learns stays with an authenticated owner, enforced by a database trigger.
- Ownership gates before any lookup. An order or booking is discussed only with the contact it belongs to. A reference that belongs to someone else is refused the same way as one that does not exist, so the reply cannot be used to probe for records.
- An audit trace of every step. For multi-step questions the plan, each lookup, each calculation and each verifier decision is stored as an append-only, versioned step. Your team can see why the assistant answered, asked a question or handed off.
- Data minimisation by design. Each step receives only the variables it needs rather than the whole conversation and knowledge base. This supports GDPR style data minimisation and access control work. It is a design property, not a certification.
- Multilingual Europe and the Americas. Compound questions are planned in English, Spanish, Portuguese, French, German, Italian, Dutch and Polish, including regional variants and code-switched messages such as Spanglish, Portunol and Franglais. Replies cover 50+ languages.
- Shared customer context. Connected channels feed a shared inbox with conversation history, cases, commitments and outcomes. Support and sales continue the same request without rebuilding its context in a separate transcript.
- Reviewed CRM and helpdesk exchange. Escalations can open tickets in Zendesk, Freshdesk or Intercom. HubSpot and Salesforce connections preview and exchange selected contacts and opportunities by explicit action. Continuous bidirectional synchronization is not claimed.
- A scoped developer surface. Business-bound API keys stored as hashes, a REST API to send a customer message and read the bound conversation, HMAC-signed webhooks, and an MCP server secured with OAuth 2.1 and PKCE.
A practical enterprise evaluation sequence
Illustrative pilot plan, not a customer case study, certification claim or throughput benchmark.
- 1. Define ownership. Choose a business, its policy owners and the staff who may change knowledge, manage integrations or handle escalations. Test representative agent permissions rather than relying only on hidden navigation.
- 2. Validate knowledge. Supply approved sources and run known-answer questions, contradictions and missing-information tests in each language you serve. Include requests that must reach a person.
- 3. Exercise gates and handoff. Ask about another customer's order, a refund that needs approval and a compound question with a missing reference. Confirm the customer sees an accurate next step and staff receive the trace.
- 4. Review integration risk. Authorize a test CRM or helpdesk, inspect the mapping, verify webhook signatures on your server and reconcile uncertain API writes before considering a resend.
Compound questions, one verified step at a time
A message that asks several things at once is split into a plan of steps. Each step runs on its own with a pruned input: an ownership check, a lookup in your records, money and time math done in code, and a final verifier that refuses any amount, promise or action that no verified fact supports. Writes such as orders, refunds and bookings never run inside a step; they pass a pre-commit check and need the customer's confirmation.
When a part cannot be answered from verified facts, the assistant asks a targeted question or hands off with the trace attached instead of guessing. The method and the current holdout safety measures are published, including privacy cases where a message points at another customer's record.
API, webhooks, MCP and SDK
- Headless Agent API. Start, continue and read conversations at https://api.octadezx.com/v1, with statuses, handoff reasons, filed actions and Server-Sent Events streaming. Keys are bound to one business and stored as SHA-256 hashes. Turns are not idempotent.
- Signed webhooks. HMAC-SHA256 signed events for escalations, orders, appointments, voicemail and refund requests and decisions, with redelivery from an outbox. Receivers should deduplicate on the event id.
- MCP server. A standard Model Context Protocol server at octadezx.com/mcp for operators who run support from Claude, ChatGPT, Gemini, Cursor or an in-house agent, with OAuth 2.1, PKCE and refresh rotation.
- Node.js SDK and Google ADK. @octadezx/sdk wraps the API, signs user tokens and verifies webhooks, and @octadezx/adk exposes a bound-conversation read tool and a human-approved message tool. Both are published on npm at version 0.1.0 under the MIT license.
- Channels. Website widget, shareable chat link, WhatsApp, Instagram, Messenger, Telegram, Discord, Slack, LINE, Viber, email and phone calls, all into one inbox and one contact history.
- In-app messengers and identity. @octadezx/messenger with React bindings and @octadezx/messenger-react-native are on npm at 0.1.0, and signed, expiring user tokens verify signed-in users. The Python SDK is on PyPI (pip install octadezx); native iOS and Android SDKs are on the roadmap.
Procurement questions deserve specific answers
- Sign in with Google is available for dashboard users. SAML SSO and SCIM provisioning are not offered today, and no SOC 2 or ISO certification, data-residency selection or verified throughput benchmark is claimed.
- Enterprise agreements include a named success manager and a written uptime commitment. No public availability percentage is published. Agree the exact support and uptime terms in writing before relying on them.
- Dashboard workflows are response guidance with existing action and handoff boundaries. They do not establish a universal transaction approval layer or automatic compliance with an internal policy.
- The ownership gates and the audit trace apply to the multi-step pipeline and the records it looks up. They support compliance work such as GDPR data minimisation; they are not a certification or legal advice.
- Agree retention, incident response, subprocessor review, permitted data types and operational ownership during your evaluation. Do not infer those commitments from a product feature list or a plan name.
Frequently asked questions
- What is an enterprise AI customer service platform? Software that answers customers across channels from approved knowledge, takes bounded actions and hands off to staff, with the access control, audit and integration surface a larger organization needs. OctaDezx provides that with per-business data isolation, role permissions, an audit trace for multi-step questions, a REST API, webhooks and an MCP server.
- Is OctaDezx only for small businesses? No. OctaDezx serves enterprise support teams, SaaS companies, mobile apps, games and e-commerce on the same platform. A larger deployment should still validate access, integration behavior and procurement requirements against the controls currently available.
- How is customer data isolated between businesses? By owner-scoped row level security policies in the database, so one business cannot read another's records through the API or dashboard. Inside a business, an order or booking is discussed only with the contact it belongs to.
- Does OctaDezx support SSO? Dashboard users can sign in with Google. SAML SSO and SCIM provisioning are not offered today. Raise identity requirements during procurement.
- Is there an SLA? Enterprise agreements include a written uptime commitment and a named success manager. No public availability percentage is published, so agree the terms in writing.
- Which languages does it support for Europe and the Americas? Multi-step questions are planned in English, Spanish, Portuguese, French, German, Italian, Dutch and Polish, including code-switched messages. Replies are written in 50+ languages, detected per customer.
- Can we audit why the AI answered the way it did? Yes for multi-step questions: the plan, each lookup, each calculation and each verifier decision is stored as an append-only step. Escalations carry the conversation and the reason the assistant stopped.
- What does the API and SDK cover? The REST API sends a customer message and reads the business-bound conversation. Signed webhooks report escalations, orders, appointments, voicemail and refund events. The server-side TypeScript SDK wraps those operations and verifies webhooks, and the MCP server covers operator tasks.
- Are security certifications included? No certification is claimed on this page. Ask about the exact certification, identity, contractual and data-processing requirements your procurement process needs before deciding whether the deployment is suitable.
Explore the connected tools
- Multi-step inquiry resolution
- Multi-step inquiry benchmark
- AI support for SaaS
- AI support for mobile apps
- AI player support for games
- Ecommerce customer care
- Enterprise rollout guide
- Customer records and CRM
- Integration availability
- Business knowledge and learning
- Platform overview
- Plans and pricing
Read more on this
- How to roll out enterprise AI customer support with clear ownership
A practical enterprise AI rollout guide: choose a bounded pilot, assign knowledge owners, review permissions, test handoffs and integrate your existing systems. - AI customer support for SaaS: billing, seats and proration done safely
How to let an AI answer SaaS plan, seat and billing questions without inventing numbers: what it needs, where math should happen in code and why proration belongs to your billing system. - Can a customer reprogram your AI support agent?
Any AI that learns from conversations is learning from text written by strangers. Here is the attack, why filtering does not close it, and what an actual fix looks like.
OctaDezx home · Platform · Pricing · AI Customer Support · AI Appointment Booking Bot · Self-Filling AI CRM · Developers & SDKs · Enterprise AI Support · Gorgias & Intercom Fin Alternative · AI Customer Service Platforms Compared · Blog · Start free trial