Wonderful
Also known as: wonderful.ai, Wonderful AI OS, Wonderful Agents, Wonderful AI Gateway, Wonderful Apps
Enterprise AI OS whose agents serve customers by voice, chat and email, run back office workflows end to end, operate legacy systems from a managed virtual machine, and deploy as SaaS, in your own cloud or on premises.
Wonderful sells what it calls an AI operating system for large enterprises. It started with customer service agents built for markets outside the English speaking world. Its live deployments include a voice agent at the telecom operator Bezeq that handles internet issues and books technician visits, and appointment scheduling for Bank Hapoalim. The platform has since widened to agents that run back office work, AI native business applications, and a gateway that governs every model call in the company.
Agents are composed from reusable skills, each packaging the instructions, tools, knowledge and checks for one task, so an improvement made to a skill reaches every agent that uses it. Business teams can build and edit agents in natural language through an Agent Builder, while engineers work with the integrations, APIs and orchestration underneath. Agents work across voice, chat, email and documents, read and write directly in CRMs, ERPs and policy systems, and can hand work to agents on other platforms over A2A. Where a system has no API, an agent can log in and work the screens itself from a managed virtual machine.
Oversight is built into how agents run. A policy layer watches every conversation and can transfer to a person or end the call, agents pause for human review and resume once approved, and every interaction is traced. Changes are tested with simulated conversations and A/B splits before they go wide, and any version can be rolled back.
Wonderful publishes a Swagger file covering hundreds of endpoints and lets customers export every agent, skill and governance setting, and the platform can run headless. Deployment runs from multi tenant SaaS to single tenant, the customer's own AWS, Azure or Google Cloud account, or an air gapped on premises install. The company holds SOC 2 Type II and ISO 27001. Each rollout starts with forward deployed engineers and deployment strategists working on site, who then hand ownership to the customer's own team.
Vendor details
Canonical URL
https://www.wonderful.ai
Category
Customer support agent
Subcategory
Enterprise AI OS for customer service and back office agents
Funding status
Private, Wonderful AI B.V., based in Amsterdam. Raised a $550 million Series C at a $5 billion valuation in September 2026, led by Insight Partners with Salesforce joining, six months after a $150 million Series B. Index Ventures, IVP, Bessemer Venture Partners, Vine Ventures and 9Yards are earlier backers. The company reports about 650 employees across more than 35 markets.
Company status
independent
Use cases & customers
Primary use cases
Target customers
Deployment options
Integrations
Agents connect directly to CRMs, ERPs, policy systems and data platforms to read and write records, and reach systems with no API by operating them from a managed virtual machine. Enterprise MCP gateways govern what agents can call, A2A hands work between Wonderful and agents on other platforms, and a published API with a Swagger file covers hundreds of endpoints. The AI Gateway takes third party tools as supported providers and connects anything else through its SDK, and a Python connector converts Wonderful agents into Azure agents.
In practice
Your customers call in a dozen languages and English first bots keep failing them. Wonderful deploys voice, chat and email agents tuned to each market, built by deployment teams who work inside your operation.
Your underwriting platform is decades old and has no API. A Wonderful agent logs in from a managed virtual machine, enters the data, runs the checks, pauses for a person to approve, then picks up where it left off.
Model spend is growing across a dozen teams and nobody can see it. The Wonderful AI Gateway routes every request through one place, caps budgets, limits premium models to the teams that need them and redacts PII on the way out.
Sources & related URLs
Related / legacy domains
Research sources
Agentic Index coverage score
12.5 / 14 capabilities · 89%
| Integrations & Tool Calling | Full |
|---|---|
|
Agents connect to the systems the work runs on, reading data, triggering actions and updating records directly in CRMs, ERPs, policy systems and data platforms. Engineers build on the underlying integrations, APIs and tools, and enterprise MCP gateways control what each agent is allowed to call. Sourcewonderful.ai/blog-articles/wonderful-an-enterprise-platform-to-turn-ai-ambition-into-agents-in-productionread 2026-10-03 |
|
| Workflow Orchestration | Full |
|
Each agent is a composition of skills and loads only the ones the current task needs, so a single agent can carry a multi step process across domains and make decisions along the way. An orchestrator and managed workflows run complete business processes from the moment work arrives until it is resolved, and work can be handed between agents on different platforms over the A2A protocol. Sourcewonderful.ai/agentsread 2026-10-03 |
|
| Knowledge Grounding & RAG | Full |
|
Every agent works from what the organization already knows, with its data, documents and policies becoming the shared context agents draw on, read under one set of permissions. Knowledge bases sit in the chain each agent draws on, and each skill can carry its own knowledge for a specialized task. Wonderful does not say publicly how documents are indexed or refreshed. Sourcewonderful.ai/agentsread 2026-10-03 |
|
| Human Oversight & Guardrails | Full |
|
A policy layer monitors every conversation for attempts to reach sensitive data or disclose what an agent should not, and teams configure the response, such as transferring to a human or ending the conversation. Agents pause where human review is required and resume once the work is cleared, and Wonderful Apps puts approve and decline controls in front of the people supervising each agent. Sourcewonderful.ai/blog-articles/the-hard-part-of-ai-agents-isn%e2%80%99t-building-themread 2026-10-03 |
|
| Security, Identity & Governance | Full |
|
Wonderful states SOC 2 Type II and ISO 27001:2022 certification, PCI DSS and GDPR compliance and EU AI Act alignment, verified independently each year, plus third party penetration tests twice a year. Permissions decide who can build, edit and ship agents down to the individual resource, and the AI Gateway sets which roles or teams may use which models. The detailed control list sits in a Vanta trust center at trust.wonderful.ai. Sourcewonderful.ai/ai-osread 2026-10-03 |
|
| Observability & Auditability | Full |
|
Conversations, decisions and tool usage are observable by default, so a team can see why an agent behaved as it did and not only what it did. Failures can be traced to the exact moment in a conversation and tagged, alerts flag tool failures and unusual volumes, and an immutable audit trail covers the environment, including every action an agent takes inside a legacy system. Sourcewonderful.ai/blog-articles/the-hard-part-of-ai-agents-isn%e2%80%99t-building-themread 2026-10-03 |
|
| Memory & State Persistence | Partial |
|
State carries through a run, so an agent that pauses for human review picks up where it left off once the work is cleared. Wonderful says every decision builds a record the platform learns from, but it does not describe a memory that agents read across separate conversations, or a scope or retention period for one. Sourcewonderful.ai/blog-articles/computer-useread 2026-10-03 |
|
| Deployment & Data Residency | Full |
|
Four deployment models are offered: multi tenant SaaS with logical isolation, dedicated single tenant infrastructure, bring your own cloud on AWS, Azure or Google Cloud with data kept inside the customer's organization, and a fully air gapped on premises install maintained on site by Wonderful's engineers. Data residency options are set out in the trust center. Sourcewonderful.ai/deploymentread 2026-10-03 |
|
| Prebuilt Agents, Templates & Packs | Partial |
|
A catalog lets a customer's own teams build an agent, skill or tool once and reuse it across the organization, and reusable skills package a procedure for any agent that needs it. Workflows such as KYC and credit servicing are taken into production by Wonderful's deployment teams, and Wonderful names no set of prebuilt agents or templates a customer adopts. Sourcewonderful.ai/agentsread 2026-10-03 |
|
| Triggers & Channel Coverage | Partial |
|
Work reaches the agents through voice, chat, email, documents and embedded experiences, with the same skills applying on every channel. Live deployments include inbound calls at a national telecom, appointment booking at a bank and collections across 2.5 million clients, and an agent waiting on a human approval resumes when the approval arrives. Wonderful does not publicly describe schedule or webhook triggers. Sourcewonderful.ai/agentsread 2026-10-07 |
|
| Model Flexibility & Routing | Full |
|
Nothing built is locked to one model, since the platform routes each task to the model that fits and new models drop in without rework. Admins keep control through the AI Gateway, which sets which roles or teams can reach which models, caps budgets and limits premium model use, and the company's CTO says customers choose the best models for each workload. Wonderful does not publicly show a model picker for each agent. Sourcewonderful.ai/ai-gatewayread 2026-10-03 |
|
| APIs, SDKs & MCP Extensibility | Full |
|
The whole platform is exposed through an API with hundreds of endpoints, published as a Swagger file, and it can run headless, so customers can build their own interfaces or use it purely as an orchestration layer. Every agent, skill, tool, governance setting and app can be exported through that API, and a CLI is part of the AI OS. The API reference sits behind a customer login at app.wonderful.ai/docs. Sourcewonderful.ai/blog-articles/open-by-default-competitive-by-designread 2026-10-03 |
|
| Testing, Debugging & Optimization | Full |
|
Changes are tested before they ship, with generative evaluations that simulate real interactions and score whether the agent met the intent behind each request, catching regressions and edge cases. In production, automated checks surface drift and failures, and A/B tests split traffic between two agent versions so the better one can be promoted and the other rolled back. Sourcewonderful.ai/blog-articles/fast-iteration-as-a-strategyread 2026-10-03 |
|
| Browser & Computer Use | Full |
|
Agents reach legacy systems with no API, logging into software such as Oracle, Epic or SAP, moving between screens, filling in fields and submitting forms the way staff do. Each one runs in a managed virtual machine with its own credentials and a controlled session, every action is logged, and a single view shows which sessions are active or waiting on human review. Sourcewonderful.ai/blog-articles/computer-useread 2026-10-03 |
|
The Agentic Index coverage score grades every vendor Full, Partial or Not documented against the same 14 buyer facing capabilities, from public evidence only. Each capability links to how all vendors in the index score on it. How this evidence is graded
Product updates
The AI Gateway from Wonderful is generally available as the control layer of its AI OS, sitting between a company's AI tools and its model providers so every model call from coding, voice and back office agents passes through one point. It reports spend by department, team, user, source, task and model with budget projections and anomaly flags, enforces per team model access and spending limits before a request reaches the model, and redacts customer data before it leaves. A router picks the most cost effective approved model for each request and moves traffic to a fallback when a provider goes down.
Bears on: Security / enterprise
View sourcePricing
Contact sales
An annual subscription, a minimum commitment, credits consumed per interaction, or a mix, set in each order form.
What is public
Wonderful publishes no price list. Its MSA does set out how billing works. Customers pay an annual subscription fee, a minimum fixed commitment, credits consumed by usage, or a combination, set in each order form. The SLA commits to 99.9% monthly platform availability, with service credits below that.
Billing mechanics
Credits are consumed as the service is used, and different interaction types consume different amounts at rates set in the order form. Usage and credit balance are visible in the account.
Cost watchouts
When the credit balance runs short, the service can be suspended or limited until more credits are bought. Automatic replenishment and low balance alerts can be set in the account.
Variable cost rationale
Wonderful meters credits, consumed per interaction at rates that vary by interaction type and are set in each order form, so spend grows with interaction volume. Use beyond the purchased allocation is billed at then current rates. Credits are prepaid. When the balance runs short, the service can be suspended or limited until more credits are bought, and automatic replenishment and low balance alerts can be set in the account. Wonderful can change consumption rates on 30 days' notice, and a change applies only to credits bought afterward.
Additional watchouts
Purchases cannot be refunded or canceled, and service can be limited when the credit balance runs out.
Overage / add-ons
Use beyond the purchased scope or credit allocation is billed at then current rates. Wonderful can change credit consumption rates on 30 days' notice, and the change applies only to credits bought afterward.
Sales call required
Yes, required for paid access
Free / trial
No free plan or trial published
Lowest paid plan
Not published
Commercial notes
Every rollout starts with Wonderful's forward deployed engineers and deployment strategists working on site, who build the first use cases and then hand ownership to the customer's team.
Key ambiguities
Subscription fees and credit consumption rates appear only in each customer's order form, so there is no public rate to compare against other vendors.
Cancellation / refund
Under the MSA, subscription and credit purchases are final and cannot be refunded or canceled, whether or not the service was used.
Support SLA / resale
The SLA commits to 99.9% monthly platform availability, measured in five minute intervals, with service credits when it is missed.
Missing data
No subscription fee, credit price or consumption rate is published.
Related vendors
- Decagon — AI agent platform purpose-built for customer support, with deep…
- 5.Y — 5
- Ada — Agentic customer experience platform whose AI agents autonomously…
- Aisera — Enterprise agentic AI platform that orchestrates specialized agents…
- ASAPP — Generative AI for enterprise contact centers that plugs into…
- Assembled — AI customer support orchestration platform that unifies autonomous…
Alternatives to Wonderful
The closest documented capability profiles to Wonderful among customer support agents tracked by Agentic Index, ordered by similarity on the same 14 point evidence the rankings use. No vendor pays for placement.
- Blits.ai12.5 / 14Fuller documented coverage on Triggers & Channel Coverage
- Yellow.ai12.5 / 14Fuller documented coverage on Triggers & Channel Coverage
- Kapture CX12.0 / 14Fuller documented coverage on Triggers & Channel Coverage
- YourGPT13.0 / 14Fuller documented coverage on Prebuilt Agents, Templates & Packs and Triggers & Channel Coverage
- ElevenAgents12.5 / 14Fuller documented coverage on Prebuilt Agents, Templates & Packs and Triggers & Channel Coverage
- Khoros12.5 / 14Fuller documented coverage on Prebuilt Agents, Templates & Packs and Triggers & Channel Coverage
Similarity is computed from each vendor's Agentic Index coverage score evidence, axis by axis, not from the totals. How this evidence is graded