An ordering bot for one restaurant brand
Takes the order on WhatsApp from menu to kitchen, checks opening hours first, and passes refunds to staff.
- Agents
- 1
- Connectors
- 2
- Channels
- 1
- Seats
- 3
Plan Launch Discuss this package
Solutions
Give a recurring job a clear owner.
Start with a result your business needs. We build and run a custom agent on Cadra, with a defined responsibility, the systems it needs and clear points for human decisions.
Build · Deploy · Maintain
Agree the responsibility before handing over the work.
Build
Make the goal concrete.
We map the job with the people who do it today. We agree the result, connect the required systems and build an agent around the process.
You contribute Real examples, access to the relevant systems and someone who can make decisions about the work.
Deploy
Decide where it can act.
We test the agent against normal work and exceptions. We set tool permissions per agent, run limits, spend caps and approval steps, then deploy it on WhatsApp, web chat, Slack or SMS.
You contribute Review of the results and agreement on what the agent may handle.
Maintain
Keep responsibility after launch.
We operate the agent, review every step in the run log, feed approved outcomes back so it learns, and tune its model routing as models and costs change. Changes in your business become changes to the way it works.
You contribute Feedback on results and notice when your rules or process change.
From intent to operationBrief · Build · Test · Run
Start with a business result worth changing.
We turn that result into a defined job, build the agent around it and keep working on it after launch. Brief. Build. Test. Run.
- 1
Brief
Agree what good looks like.
Show us where work stalls, what a good result looks like and which decisions belong to your people.
- 2
Build
Give the agent a job it can finish.
We load a knowledge base for your business, connect your systems through connectors and MCP, and set tool permissions and handoff points for each agent.
- 3
Test
Prove it on your real cases before go-live.
We run the agent on your real examples and awkward cases, and agree with you what needs human review before it goes live.
- 4
Run
Keep it useful as the business changes.
We maintain the agent, review its run log, A/B test models on real traffic and adapt the process as the business changes.
Where work can move
Look for the work that keeps coming back.
Campaigns, questions, credit checks, candidate screening, follow-ups. Start where the result is clear and routine work keeps pulling people away from it.
Growth marketing
Bring the marketer to the merchant.
YoboLabs onboards merchants on WhatsApp, drafts their campaigns and sends a brief every morning. Case study Read the caseCustomer service
Keep the conversation moving when a person steps in.
YoboLabs support lets an agent and a person take turns in the same thread, with context retained when the agent resumes. Case study Read the caseStaff assistants
Bring business information into the conversation.
YoboLabs includes a staff assistant over WhatsApp, Slack and web, plus daily AI briefs. Built on Cadra Discuss staff assistanceTalent
Find the people worth hiring.
JetDevs screens resumes, proctors tests, scores interviews and matches engineers with AI. Case study Read the caseSales and CRM
Make follow-up a defined responsibility.
A possible starting point: an agent with a clear follow-up goal and rules for when a salesperson takes over. Application to discuss Discuss sales follow-upBack-office work
Put the process into motion.
Cadra boards run procedures lane by lane, with agent lanes for the routine and human lanes for decisions and approvals. Cadra capability Explore CadraCredit and fintech
Decide credit in seconds, with reasons.
SoyakaAI assesses an application, matches an offer inside the lender's rules and explains the outcome. Case study Read the caseProof of completion
Reward work that was really done.
ChainXP writes missions with AI, audits the evidence and certifies completion. Case study Read the caseFrom build to operation
An agent needs more than a brief.
The package defines the job and the work around it. Build, integration, testing, deployment and ongoing operation are planned together.
- A defined responsibilityThe goal, the board it runs on and the lanes where a person approves or takes over.
- Connections to your systemsConnectors to your systems, including MCP, plus custom connectors we write for the build.
- A place to testA sandbox where the agent runs your real examples and edge cases before release, with every step in the run log.
- Agreed channelsWhatsApp, web chat, Slack or SMS, with one memory across them, wherever your customers or staff already are.
- Controls for the jobTool permissions per agent, approvals at the steps you choose, and cost, time and tool-call limits on every run.
- Ongoing operationWe run it after launch: model routing tuned, learning from outcomes reviewed, and a record of every change.
Start with a defined package. Add responsibility deliberately as your needs change.
Choose a starting pointBy use case
One clear responsibility. A defined package.
Begin with the work you want an agent to own. Agree the goal, systems, channels and human decisions before the build begins.
Support on web chat and WhatsApp, with handoff to your team
One agent answers questions, one looks up orders and returns. Hard cases go to your staff in Slack with the whole thread attached.
- Agents
- 2
- Connectors
- 4
- Channels
- 3
- Seats
- 8
Plan Grow Discuss this package
A loan operations desk
Several agents read applications, check documents against your credit rules and draft credit memos. Every decision waits for an analyst. Deployed privately.
- Agents
- 8
- Connectors
- 6
- Channels
- 2
- Seats
- 30
Plan Scale Discuss this package
Sales follow-up for your team
Reads new leads from your CRM, drafts the follow-up, books the meeting, and reports in Slack.
- Agents
- 1
- Connectors
- 3
- Channels
- 1
- Seats
- 5
Plan Launch Grow, if every email needs a person's yes before it goes out Discuss this package
PlansScoped by agent
Three plans, scoped by agent. Everything is included.
Every plan is a fixed package scoped to the agents you need: a defined build, then a monthly run over a minimum term. Nothing is billed by the hour, and nothing is left open-ended.
Launch
3-month minimumOne agent doing one job. The smallest way to put an agent to work.
- Agents
- 1
- Connectors
- 2
- Channels
- 1
- Runs a month
- 2,000
- Procedure changes a month
- 1
- Operator seats
- 3
- Approvals and handoffs
- —
Grow
6-month minimumA few agents that work together, with a person approving the steps that matter.
- Agents
- 3
- Connectors
- 6
- Channels
- 3
- Runs a month
- 6,000
- Procedure changes a month
- 3
- Operator seats
- 10
- Approvals and handoffs
- Yes
Scale
12-month minimumA desk of agents running a whole operation, with the controls a regulated business needs.
- Agents
- up to 10
- Connectors
- 20
- Channels
- 10
- Runs a month
- 20,000
- Procedure changes a month
- 10
- Operator seats
- 25
- Approvals and handoffs
- Yes
- A service-level agreement on response times
- A staging copy to test every change before it goes live
- Option: private deployment, including in China
- Option: bring your own AI model keys
Plans are scoped to the agents you need. We quote after a short call: no day rates, no hourly billing, no open-ended scope.
Per agentWhat one agent includes
What counts as an agent
An agent does one job: it follows one procedure from start to finish, such as taking an order or checking a loan application. Every agent comes with the allowance below. Allowances are shared across all your agents.
- Job
- 1
- one procedure
- Connectors
- 2
- links to your systems
- Channel
- 1
- where people reach it
- Runs
- 2,000
- a month
- Procedure changes
- 1
- a month
- A runOne piece of work done start to finish: one conversation answered, one order taken, one application checked.
- A connectorA link to one of your systems, such as your point of sale, CRM, helpdesk or document store.
- A channelOne place people reach the agent: website chat, WhatsApp or Slack.
- A procedure changeA new rule or step, such as a new refund limit. Keeping the agent working as agreed is never a change; that is part of running it.
Who needs a seat
Needs a seat
People on your team who work with the agent: they approve steps, review the run log, take handoffs and edit the procedure.
Never a seat
Your customers, and anyone else who talks to the agent. They are covered by runs, however many people that is.
Add-onsAs you grow
Need more? Add exactly what you need.
When you outgrow an allowance, you add the piece you need for the rest of your term. We agree each one with you before it starts.
- + AgentOne more agent for one more job, with its own allowance. Grow and Scale only.
- + ConnectorOne more link to one of your systems, built by us.
- + ChannelOne more place people can reach the agent.
- + Run pack2,000 more runs a month.
- + Operator seatOne more team member who approves, reviews or takes handoffs.
- + Procedure changeOne more change to the procedure in a month.
Launch covers one agent; a second agent means moving to Grow.
Questions before you delegate
Know what you are handing over.
What is a good first job for an agent?
A recurring job with a clear result, accessible information and rules your team can explain. Ordering, support and internal procedures are useful places to look. We define what the agent can finish and when it should involve a person.
How soon can we see something working?
We agree the timeline around the job, its connections and the testing it needs. Our WhatsApp growth marketer went from spec to production in 21 days, and reached a real phone on day 6. Every build is different; that one is on the record.
Which models can we use, and how do you run them?
Every skill, role and agent can run on its own model. Routing sends each task to the model that fits its type, cost and speed. We A/B test models on real traffic, and balance load across AI providers with automatic failover. Closed hosted models, open-source models such as Llama, Qwen, DeepSeek, Kimi and GLM, and local models through Cadra desktop work side by side, and you can bring your own keys.
Can a model run locally? What about China?
Yes, local models are an option. For a China deployment, we work through model availability, the systems it connects to and where data will travel. Those details belong in the deployment plan. Choosing a local model alone does not determine where every part of the service handles data.
How do we control the agent?
You define its goal and authority. We set tool permissions per agent, the steps that wait for approval, and cost, time and tool-call limits on every run, with spend caps on top. The run log records every step, tool call and cost, so you can see what happened and decide what should change.
What happens if an answer isn't right?
We review the run in the run log, correct the cause and test the change before it goes back live. Approval steps and human handoffs define where people step in. In our support build, a person can take over the same conversation, and the agent stops replying once they do.
Can we start small?
Yes. Start with one clearly defined responsibility in a packaged engagement. Agree the boundaries and the result to review before adding more work.
What does it cost?
Plans are packaged per agent: a fixed scope, built for you, then a monthly run over a minimum term. The build, the connectors we write for you, launch into your channels and running it after launch are all included. We quote after a short call about the work you want to hand over.
Do our customers count as seats?
No. Seats are only for people on your team who approve steps, review the run log, take handoffs or edit the procedure. Everyone who talks to the agent is covered by runs.
Do you work on-site?
We work alongside your team through every stage, from the first brief to running the agent after launch.
Who owns and controls our data?
Data ownership and permitted use are set out in the agreement. Cadra separates clients into isolated workspaces. Before the build, we agree what information the agent can access and where the selected model and connected systems will process it.
Will it work with our existing software?
Connecting the agreed systems is part of the build. Cadra connects to business software through connectors and MCP, with permissions per tool action, and we build custom connectors where needed. We check access and what each system allows before committing the integration to the package.
Do we need an internal AI team?
Imaginato handles the agent's build and operation. Your team supplies the business knowledge and decides what it may do. You need someone responsible for the process and its results.