A separate bot for every team
Voice, chat, coaching, QA and back office each buy their own tool, with their own rules and their own logs.
The AI-first platform for regulated operations
One platform. Build once, every agent inherits it.
Spine is the shared foundation for AI agents: connectors, orchestration, skills, governance and audit, built once. Every agent on it follows the same rules and leaves the same record.
Live proof
Numbers from the running platform.
- Agents registered
- Loading
- Governed actions, last 30 days
- Loading
- Approval decisions, last 30 days
- Loading
- Guardrail blocks, last 30 days
- Loading
- Frameworks with controls mapped
- Loading
- Controls mapped
- Loading
- Connected systems reporting live
- Loading
These are live platform aggregates, anonymised across tenants. "Not yet public" means the number is below our publishing floor.
The problem
Every team buys its own bot. Nobody owns the rules.
Separate bots mean separate rules, separate logs and no single answer when a regulator asks what the AI did. The cost grows with every new tool.
One fix, fifteen times
When a rule changes, every bot needs its own fix. Miss one and it keeps doing the old thing to customers.
Nobody can prove what the AI did
Regulators and boards ask who approved an action and why. Scattered logs and screenshots do not answer that.
One platform, three streams
Build once. Every stream inherits it.
Three lines of business run on the same foundation. A compliance fix happens once, not fifteen times. Every new agent starts governed.
Voice AI plus CX and Collections
AI takes tier-one service and recovery work. People keep the complex negotiations, disputes and regulatory cases.
Revenue focus- Voice AI, LiveAssist, NEQQO and Collections
- FDCPA, Reg F and TCPA rules checked before every contact
- Eval-gated before an agent goes live
Internal Efficiency
Governed agents for the work inside the company, from engineering to finance to the leadership desk.
- Fleets of coding agents in a governed IDE
- Workforce management, finance and reporting
- A second brain for every leader
Bespoke Solutions
The same gates and the same ledger, pointed at a new problem. Built on Spine like any other agent.
- Trust and Safety and content moderation
- Fraud, AML, KYC and compliance operations
- Data curation and annotation
Spine foundation
- Connectors
- Orchestration
- Skills
- Governance
- Audit
Built once and shared by every agent in every stream. One control plane owns the rulebooks, guardrails, approvals and audit.
How Spine works
Build it, run it under gates, prove what happened.
The same three steps for every agent in every stream. Each one is something your risk team can check for itself.
Build
Write the rulebook in plain language in Agent Studio: what the agent may do, must never do and how it is tested. Try it on sample calls before it goes live. Agent Studio
Run under gates
Guardrails check every step. Routine actions clear at once. Big ones wait for a named person, so no approval means no action. Approval Gates
Prove
Every step, tool call and decision lands on a hash-chained ledger. Change any past entry and verification shows where. Audit Ledger
One engine, many places
One brain owns the rules. Many hands do the work.
The same engine runs wherever the job is, and each copy enforces the rules locally. Customers never wait on a round trip to check a rule.
One brainThe Spine control planeRulebooksGuardrailsApprovalsAudit
Rules flow downrulebook v7
Records flow upSigned
Chat and officeCentralChecks rules locally
Live voiceAt the edgeChecks rules locally
ReportsBatchChecks rules locally
Proactive agentsSandboxedChecks rules locally
At the top sits one brain, the Spine control plane, holding the rulebooks, guardrails, approvals and audit. Below it are four hands, the same engine placed where each job runs: chat and office work centrally, live voice at the edge, reports as batch work and proactive agents in a sandbox. Rules flow down to every hand with a version stamp. Signed records flow back up to the brain. Each hand checks the rules locally.
Chat and office work, central
Assistants for mail, documents and chat run beside the control plane.
Live voice, at the edge
Voice agents run close to the call, so a turn takes under a second.
Reports, as batch
Report building runs as scheduled batch work through one governed pipeline.
Proactive agents, sandboxed
Agents that act on their own run walled off, with only the tools they are allowed.
Record first, then act
Every action is written to the record before it runs, then synced to the tamper-evident ledger. No action without a record.
Routine actions never wait. Big ones always get a human.
Each engine enforces the rules locally, so pre-approved routine work clears at once. Anything bigger pauses for a named person.
Our promises
Two promises we build into the platform.
The bot is not the difference. The domain logic, compliance, quality and reporting around it are, and they stay under your control.
Your data and workflows stay yours.
Rulebooks are plain text you own and can take with you. Each client sits behind its own database walls, agent memory included. Open-weight models can run in your region.
A 72-year track record of regulatory compliance, extended by design.
Compliance as code: guardrails the model cannot override, a contact clock that fails closed, consent kept as evidence and a person deciding every big action.
Inside Spine
What your risk team and operators see.
The ledger shows every step and whether the chain still holds. The Cockpit shows every agent and lets an operator pause or cap it.
Audit ledger
Each entry carries the hash of the one before, so tampering shows.
Cockpit
Pause a product, set an autonomy ceiling or cap daily spend, with a reason on record.
Why one platform
Point tools vs Spine.
Good point tools each solve part of the problem, then leave you to stitch the rules and logs together. Spine governs every agent in one place.
| Capability | Voice vendor | QA vendor | GRC tool | Spine |
|---|---|---|---|---|
| Governs the agents that talk to customers | PartlyIts own agents only | No | PartlyPolicies, not the live agent | Yes |
| Approval before a write | PartlyHandover to a person | No | No | YesFails closed |
| Tamper-evident audit | PartlyCall logs | PartlyRecordings and scores | PartlyEvidence records | YesHash chained, verifiable |
| US, Canada and India rules in one control set | No | PartlyScorecards per rule | PartlyFrameworks, not frontline rules | Yes |
| Live coaching of human agents | No | PartlySome offer it | No | YesLiveAssist |
| Conversation QA | PartlyBasic call analytics | Yes | No | YesNEQQO |
| Pause a product in one click | PartlyIts own agents only | No | No | YesFrom the Cockpit |
Columns describe typical tools in each category, not any one vendor. Partly means some tools in the category do some of it.
Regulation
US, Canada and India rules in one control set.
Map a control once and see where it applies in each market. Evidence is collected from the platform itself.
United States
Contact limits, consent and privacy rules for lending, collections and servicing.
FDCPA and Reg FTCPAGLBACanada
Federal privacy rules for how customer data is collected, used and kept.
PIPEDAIndia
Data protection and incident reporting rules for digital lenders and service operators.
DPDPCERT-In
Controls mapped:NIST AI RMFSOC 2ISO 42001
Controls mapped and evidence collected. These are not certifications.
Deployment
Your region, your models, open weights.
Choose where Spine runs and which models it calls. Your data need not leave your region, and the gates and the ledger stay the same wherever it runs.
Your region
Run where your regulator expects your data to live. Each client is walled off in the database itself with row level security.
Open weights
Open-weight models are supported. Voice AI already runs on open-weight models on Google Cloud in Mumbai.
Your models, one router
Agents are built on Google ADK and governed by Spine. A model router picks per job, with fallbacks and a fuse per provider.
Questions
Frequently asked questions.
Spine is one AI-first platform, built once and inherited by every agent. It holds the connectors, orchestration, skills, governance and audit, so each new agent starts with the same rules and the same record.
Voice AI plus CX and Collections covers tier-one service and recovery, with live guidance for human agents. Internal Efficiency covers coding agents, operations and a second brain for leaders. Bespoke Solutions covers Trust and Safety, fraud and compliance operations and data annotation. All three run on the same foundation.
You do. Rulebooks (AOPs, the operating procedures an agent follows) are plain text you own. Rulebooks and the governance contract export as plain files. Each client sits behind its own database walls, agent memory included.
A model router picks the model per job, with fallbacks and a fuse per provider. Open-weight models are supported. Voice AI already runs on open-weight models on Google Cloud in Mumbai. Agents are built on Google ADK and governed by Spine.
Pre-approved routine actions under the day's running total clear at once and are recorded. Anything else pauses for a named person while the conversation keeps flowing. The facts are checked again, then it acts. If nobody decides, nothing runs.
The same engine is placed where each job runs: chat and office work central, live voice at the edge, reports as batch and proactive agents sandboxed. Region is agreed per deployment.
Agents pin the contract, never the engine. The engine keeps moving and still honours older contracts, like phone apps after a system update. A gate refuses to promote an agent the engine cannot honour.
Start with one workflow. Email hello@billgosling.ai and we will walk through it on Spine, agree the rules it must follow and scope a first use case and its price with you.
The question is not whether. It is how fast.
Bring one workflow. We will run it on Spine end to end: the agent proposes, a person decides and the ledger proves it.