Create an AI-Native Service Platform
Put your firm’s expertise into an application: agents do the work, your qualified people review and sign, and you can show what happened.
- Your method, carried out by agents
- Qualified people review and sign
- A record of what happened
Sell the outcome. Keep the expertise.
An AI-native service firm, in law, accounting, brokerage or support, sells finished work rather than seats. Agents carry out the firm’s own method, and the firm’s qualified people review and sign what leaves it.
Wildo is the backbone of that firm’s application. Every agent acts under a named authority, every decision reaches someone entitled to take it, every correction an expert makes is kept as the firm’s know-how, and the firm runs the application on its own infrastructure.
From a request to the proof of what was done
Agents do the work
Agents run inside the application with the firm’s instructions and tools. They can search the firm’s records and documents without seeing more than the person they work for may see, and every change they make is recorded as the agent’s, under the authority it acted for.
A qualified person decides
When an agent reaches a step that needs judgement, the work pauses and the decision goes to the reviewer the step names: a named person or anyone holding a role, never the person who asked. The reviewer approves, amends or rejects, and each correction is recorded, so the firm learns where its agents need its experts.
The firm can show what happened
Every action is written to an audit trail that can be sent on to a security team’s SIEM. Each model call is attributed to the client and the case it served, so the firm knows what a case cost. Data is kept for the periods the firm declares, held when a matter requires it and deleted on schedule.
Example: A bookkeeping firm closing its clients’ months
Each client has its own workspace in the firm’s application, and the client decides which of the firm’s experts and agents may work there. An agent prepares the month-end reconciliation from the client’s records and proposes adjustments for the entries it cannot match. A qualified accountant reviews each proposal, amends what needs amending and approves it. The adjustment is recorded with the agent that proposed it and the person who approved it, the correction feeds the firm’s view of how often that agent is overridden, and the firm sees what the close cost in model calls.
For engineers
Available today
Each of these has run end to end in a live application:
- Agents defined with the firm’s instructions and tools, with retrieval over its records and its documents, always within the caller’s own access.
- Agents that run on a schedule, under a machine identity each organization grants, pausing for a reviewer when they reach a gated step.
- Attribution on every audit record: which kind of authority acted, and which agent, conversation and tool did the work. Exported to a SIEM as CEF, LEEF or OCSF, and checkable against an independent ledger.
- Review gates on an agent’s actions: a step the requester confirms, or a decision routed to a named person or a role, which the requester can then never take. A reviewer inbox, approval with modifications, and the requester’s authority checked again at approval.
- Review deadlines, reminders and escalation, counted in business hours against the firm’s calendar. A missed deadline never approves or rejects anything.
- A record of how experts corrected the agents: every approval, amendment and rejection, with the firm’s own reason codes and the override rate per agent, tool or case. The content of corrections is kept only if the client chooses to keep it, and it can leave as an evaluation dataset only on a recorded legal basis, pseudonymised, with the clients that opted out left out.
- Relationships the client controls, which let the firm’s named experts work inside the client’s organization. The client approves each expert, admits the firm’s agents it wants, can grant named administrative rights for a limited time, and applies its own sign-in policy when an expert enters. Every change an expert makes there is in the client’s own audit trail.
- Cost per case: each model call is attributed to the client and the case, with monthly totals a billing manager can link to an invoice.
- Deliverables generated from templates, filled from the case’s own records.
- Signed inbound webhooks, stored durably, delivered in order for each declared ordering key, retried and replayable by an administrator.
- Retention periods bound to each kind of data, with deletion on schedule on both supported databases and a record of every scheduled deletion.
- Legal holds that suspend every deletion of the held data inside the application, and statutory retention inside an erasure request: what the law requires is kept, restricted, until its period ends.
Landing now
These are built and tested, and are completing their first run in the conditions they are meant for:
- Agents started when a message arrives, under the same grant and review gates as scheduled ones.
- Receiving email into the application, routed to the right organization or client mailbox, with attachments scanned before anything acts on them. It has run end to end against a stand-in for the email provider; the first run with the real provider is next.
- Showing a record’s message threads in the application, and replying from it.
- Reading what clients send: the text inside their documents and scanned pages, extracted for agents to work on.
- The firm’s own accounting or CRM system as a system of record: Odoo, SAP S/4HANA, Salesforce, Microsoft Dynamics 365, Business Central, NetSuite or any OData service, read in place, copied and kept current, or written back under the system’s own rules.
Where it is going
The direction is a platform on which a firm runs its whole delivery through agents and its experts:
- Production hardening. Ending a relationship or a role takes effect at once on live connections and queued work, not only on the next request; experts serving a large number of client organizations; a legal hold that also suspends the erasure effects held at outside vendors; first runs with a real SIEM, outside MCP and A2A clients and a cloud document-reading service.
- Deeper case work. Deadlines and lapse on long-running agent workflows, not only on conversational ones; retention periods counted from the closure of a matter or a case; data read from every supported accounting and CRM system confined company by company, as it already is for Odoo.
- A creation line for service firms. Today a firm’s application is specified and built with us. The factory will carry the firm’s archetype itself: cases, review stages, client portals and agents as a reference structure the coding agents start from, so a new firm’s platform begins from the method rather than from a blank application.
How it is built with you
The firm’s experts bring the method: the criteria, the decision points and who is qualified to sign. Together we put it into the application’s specifications, its agents and their tools, and its review gates. The result is a repository the firm owns and runs with Docker Compose or Kubernetes on its own infrastructure.
Go deeper
The parts of Wildo this page draws on, each explained on its own page.
- AI assistants & task agents Application agents with their own instructions, tools and structured or conversational outputs.
- AI, retrieval & agent protocols Build assistants that retrieve knowledge, call application tools and interact through agent protocols.
- Identity, organizations & access Manage users, customer organizations, roles and service identities across application operations.
- Audit trails & data privacy Record accountable actions and handle personal-data requests, retention and erasure through application services.
- Providers, OAuth & external data Call provider operations, connect customer accounts, and read, write back or synchronize the ERP and CRM a company already runs.
- Docker & Kubernetes deployment Generate deployment configuration for your application services and their backing infrastructure.
Your expertise is the product.
The application is how that expertise reaches your clients, reviewed by your people, improved by their corrections and recorded for anyone who needs to check.
Apply for the private beta.
The private beta opens with the first release, and we are choosing the companies that take part. Tell us what you want to create: we will say plainly whether it fits and what happens next.