Platform
One platform for agents, controls and evidence
Forge has five layers. Your teams use the agents and compliance packs; governance, the improvement loop and the runtime keep those agents within the limits your organisation sets, from pilot to production and after launch.
How the layers fit together
Service, operations and oversight agents arrive with sector prompts, tools and test sets. You configure them for your systems and languages.
A pack turns a set of rules into agent behaviour, automated checks and the records each check produces. Coverage differs by country and sector; we confirm it for your use case.
A policy engine decides what an agent may do and when a person must approve. Every answer and action is logged so it can be reviewed later.
When your reviewers correct an agent, the correction is kept. A new version is promoted only if it passes the same agreed tests.
An in-country cloud region, a sovereign or national cloud, or your own servers with no internet connection.
Selected: Governance and assurance
Agent catalogue
Agents for service, operations and oversight
Catalogue agents are configured for your systems rather than built from scratch. Availability varies by sector and country.
Service
- Customer service by voice and chat
- Agent assist
- Call review and quality
- Complaints handling
Operations
- KYC (know-your-customer) and onboarding
- Claims intake
- Collections
- Document processing
Oversight
- Compliance monitoring
- Audit evidence
- Model validation reports
- Risk alerts
Governance and assurance
Every answer and action, on the record
A policy engine decides what an agent may do and when a person must approve. Every answer and action is logged so it can be reviewed later.
Audit log
Audit log · today
Illustrative · sample data| Time | Event | Record | Detail | Actor |
|---|---|---|---|---|
| 09:16:40 | report.generated | VR-0817 | Validation report for call review v8 | system |
| 09:14:38 | action.held | REQ-5530 | Card switch held by rule BNK-MS-03, awaiting review | policy |
| 09:14:37 | check.flagged | CR-2291 | Possible mis-selling at 00:37 | call-review-v8 |
| 09:14:36 | check.passed | CR-2291 | Disclosure, identity questions, complaint logged | call-review-v8 |
| 09:12:44 | policy.decision | COL-4471 | Payment plan 4,200 within limit 5,000: allowed | policy |
| 09:12:40 | pii.mask | COL-4471 | Account number and national ID masked before the model | pii-guard |
| 09:10:02 | model.promoted | call-review-v8 | Promoted after validation sign-off | model-risk |
| 09:05:11 | egress.blocked | NET-0092 | Request to a non-approved region refused | gateway |
Improvement loop
A new version ships only when it passes the same tests
When your reviewers correct an agent, the correction is kept. A new version is promoted only if it passes the same agreed tests.
Validation / Call review agent
Release candidate v8 · call review
Illustrative · sample data| Test | Result | Threshold | Against threshold | Status |
|---|---|---|---|---|
| Disclosure accuracy | 98.6% | ≥ 97% | Passed | |
| Identity questions asked | 99.1% | ≥ 98% | Passed | |
| Mis-selling detection | 94.1% | ≥ 92% | Passed | |
| Complaint capture | 96.8% | ≥ 95% | Passed | |
| Escalation correctness | 96.3% | ≥ 95% | Passed | |
| Unsupported answerslower is better | 0.4% | ≤ 1% | Passed |
Compared with v7: mis-selling detection +1.8 pts. Promotion needs Model risk and Compliance sign-off.
Validation / Reports / VR-0817
Validation report · call review v8
Illustrative · sample data- §1Scope and intended usePost-call conduct review, retail cards, two languages
- §2Data1,200 anonymised calls from the bank’s own archive
- §3Results6 of 6 agreed thresholds met
- §4LimitationsNot validated for business-banking calls
- §5ControlsHuman approval on product switches; stop switch tested
How it works
From pilot to production
Configuration, testing, validation, your approvals and production are separate steps. How long the approvals take is up to your organisation.
- 1
Agree the pilot
One use case, the data, the environment and the success criteria, in writing.
- 2
Configure
Choose a catalogue agent, apply the relevant compliance pack and connect test systems.
- 3
Test and validate
A working agent in your test environment is scored against the agreed tests on your data.
- 4
Your approvals
Your risk, compliance and security teams review the evidence on their own timetable.
- 5
Production with controls
Approvals, limits and audit records stay on. Updates ship only when they pass the same tests.
Next step
See Forge on a use case like yours
See your use case in a demo. If it fits, we agree a pilot using your approved data and success criteria.
Before a pilot begins, we agree
- One initial use caseA single, well-defined task to start with.
- Data and environmentWhich data the agent may use and where the pilot runs.
- Success criteriaThe tests and thresholds that count as a pass.
- Evaluation deliverablesValidation results, audit records and a validation report.
- The decision processWho reviews the evidence and decides on production.
Pilot scope and fees are agreed before work begins.