Skip to main content
connic

Agents for regulated work,
built like your other software.

Write agents in YAML and code, release them through pull requests and tests, and keep a person on every risky step. Your data protection officer gets clear answers on where data goes.

agents/dispute-intake.yaml
+4 -0
tools:
- core_banking.find_card_payments
- cases.open_dispute
- payments.provisional_credit
output_schema: dispute_case
approval:
tools:
- payments.provisional_credit: param.amount_eur > 250
connic/pr-testsAll tests passed
dispute-intake
Run started Oct 30, 08:14
Awaiting Approval
payments.provisional_credit
case_id
"DSP-30418"
amount_eur
412.80
ApproveReject

Hand back-office work to agents, with people on the risky steps.

Examples for a retail bank's operations team. Each agent is a file in code, each risky action waits for a person, and every decision is on record.

At 06:00 the agent groups the night's rejected transfers by cause and proposes a fix for each. Nothing is resubmitted until the payments team approves it.

Payment exceptions
ops.example.com/payments/exceptions/2026-10-30
OperationsMS
Overnight exceptions, Oct 30
31 rejected transfers · sorted at 06:04 by payment-exceptions
2 need approval
CauseCountProposed fix
IBAN check failed14Resubmit 11 with the IBAN from the customer’s payee list, ask 3 customers to confirmApprove
Received after the cut-off9Resubmit in today’s first runApprove
Held by fraud rules5Left for the fraud teamNo action
Receiving account closed3Return to sender, letters draftedReview drafts

Nothing is resubmitted until someone on the payments team approves it. Read from the payments database at 06:00.

An advisor asks in the internal portal. The agent answers from your policy documents with section and page, and the advisor decides what to tell the customer.

Ask · Advisor Portal
portal.example.com/ask
Advisor PortalLW
Recent questions
Joint account for a minor
Card limit after a move abroad
Overdraft on a basic account
Can a customer with a basic account get an overdraft of €500?

Not on the basic account. Overdrafts are only offered on current accounts with a regular income (section 3.2).

The customer can switch to a current account first. An overdraft of up to €1,000 then needs the standard credit check (section 3.4).

Credit policy 2026 · 3.2 · p. 14Credit policy 2026 · 3.4 · p. 16
Ask about a product or policy

A complaint arrives by email. The agent sets the reply deadline and drafts an answer from the case history, and the complaints team edits and sends it.

CMP-7712 · Complaints
ops.example.com/complaints/CMP-7712
OperationsAH
Balance not paid out
CMP-7712 · Murat Yilmaz · Oct 29
murat.yilmaz@example.net

I closed my savings account on 12 September and still haven’t received the balance of €2,318.40. Nobody has answered my two calls.

Category
Account closure, payout delayed
Reply due
Nov 20
Found
Closed Sep 12, no payout account on file
Draft replyNot sent

Dear Mr Yilmaz,

thank you for your letter of 29 October, and I’m sorry you had to wait. Your savings account was closed on 12 September, but we had no account on file to pay the balance into.

Please send us the IBAN you would like us to use, and we will transfer €2,318.40 within two business days.

Edit draftSend reply

Less to build, less to explain in every review.

Teams that build agents on their own stack maintain servers, queues and test harnesses, then answer the same data protection questions for every new agent. On connic, the answers live in the project.

Your own stack
Your agents in YAML and code
On connic
Your agents in YAML and code
connic

Servers, queues and retries

Your own stack: Servers, queues and retry code for each agent

On connic: connic builds and runs the agents your team writes in YAML and code, and retries failed model requests

Report what the agents did and what people decided.

Every run and every approval is recorded, so your operations and risk reports draw on the same data.

  • Cases per agent

    Runs per agent and per case type over time, on a dashboard.

  • Decisions by people

    Every approval request with its decision, who made it and when, in connic and in the audit log.

  • Processing time

    How long runs take on each agent's page, on average and for the slowest ones.

  • Cost per case

    Model cost grouped by a value such as the case type, on the usage page.

Clear answers for your data protection review.

Each stop of a run is a setting your team controls. connic/* model calls run in the EU, and a request fails rather than leave it. You contract with connic GmbH in Munich under a data processing agreement.

Providers you can name
The model catalog marks EU-native providers and any provider that keeps request data for 30 days. Managed inference isn't used to train models.
Core systems stay in your network
A bridge inside your network opens only outbound connections, and an allow list limits which hosts it can reach.
Less personal data in run records
Redact fields such as IBANs from stored runs and shorten how long each environment keeps them.
Evidence on request
The audit log records changes and approval decisions. On enterprise, AI Governance exports evidence snapshots for your auditors.
RP-0117 · Data Protection
privacy.example.com/records/RP-0117
Data ProtectionJO
Payment dispute intake
Record of processing · RP-0117 · Owner: Operations
Approved Oct 28
Dispute emailsBank mail server, read through a connic bridge
Model callsconnic/mistral-large-2512 in the EU, EU-native provider, not used for training
Transaction lookupCore banking, through the same bridge, one allowed host
Credits over €250A case handler decides, recorded in the audit log
Run recordsconnic project, IBANs redacted, deleted after 30 days
Processor: connic GmbH, Munich. Data processing agreement accepted Sep 12. Reviewed by J. Okafor, Data Protection.

Every change goes through review and tests.

The dispute agent from the top of this page is one YAML file with a test suite next to it. A deploy goes live only when every case passes.

Checks on pull requests
On GitHub and GitLab, the suite runs on each pull request and reports connic/pr-tests as a commit status, so the merge can wait until it passes.
A gate on every deploy
A push to the branch mapped to Production builds the deployment and runs every case. Until all of them pass, the current version keeps serving.
Rollback without a rebuild
Activate an earlier deployment, and it serves again with the connectors it declared.
agents/dispute-intake.yaml
version: "1.0"
name: dispute-intake
description: "Turns dispute emails into cases"
model: connic/mistral-large-2512
fallback_model: connic/mistral-medium-3.5-128b
system_prompt: |
  Read the customer's email and find the disputed card payments.
  Propose a provisional credit only when policy D-4 allows it.
tools:
  - core_banking.find_card_payments
  - cases.open_dispute
  - payments.provisional_credit
output_schema: dispute_case
approval:
  tools:
    - payments.provisional_credit: param.amount_eur > 250
tests/dispute-intake.yaml
version: "1.0"

tests:
  - name: double_charge_waits_for_a_person
    payload: '{"subject": "Charged twice for my hotel booking"}'
    mocks: core_banking
    approval_decisions:
      - tool: payments.provisional_credit
        decision: approve
    expected_tool_calls:
      - payments.provisional_credit: invocations == 1
    expected_result: output.case_type == "duplicate_charge"

Frequently Asked Questions

Agents are YAML files, and their tools, middleware and tests live next to them in your code. Changes go through your normal review, and connic builds, deploys and runs them from a connected Git repository or with connic deploy. The dashboard is where you inspect runs and manage settings.

connic/* model calls always run in the EU. Everything else follows your configuration: models with your own provider key, the systems your tools call, guardrail and judge models, and your connectors' destinations. The project stays in the EU only when each of those does. The sub-processor list names where connic processes data.

Managed inference is not used to train models. For billing, connic records only the model and token counts of each call, and the Privacy Policy lists the narrow security and error-retention exceptions. Providers you bring your own key for handle data under your terms with them.

You choose which actions need approval, for example credits above €250. The run pauses until someone with approval rights decides in connic, through the REST API or in your own app, and the audit log records the decision, who made it and what the agent wanted to do.

Yes. Add your key once per Project and pick the provider's model in the agent file. Those requests go to the provider under your own contract and fall outside the connic/* EU guarantee.

No. On enterprise, AI Governance helps you document AI systems, assessments, controls and incidents and export evidence for review. Classification, legal advice and certification stay with you and your counsel.