Rulebooks you own
Each agent runs on a rulebook, called an AOP: what it may do, what it must never do and how it is tested. It is plain text, owned by the team that runs it.
Our promises
Two promises behind every agent.
Your data and workflows stay yours. Compliance is written as code, checked before every action and proven after it. Both hold for every agent that runs on Spine.
Promise one
Your data and workflows stay yours.
Your procedures are the asset, not the bot. We enforce them, record every action and never let an agent act outside them. You keep the procedures.
Portable by design
Rulebooks and the governance contract export as plain files. If you leave, your procedures leave with you in a form any team can read.
Walls around every client
Each client sits behind its own database walls, enforced by row-level security in Postgres. Agent memory sits behind the same walls.
Models in your region
Open-weight models can run on inference in your region, so your data need not leave it. The model router picks the model and keeps a fallback ready.
Why this matters
The value we add is domain logic, compliance, QA and reporting. None of it depends on locking your data in. Rulebooks and the governance contract export as plain files today; other records follow your contract and deployment.
Promise two
A 72-year track record of regulatory compliance, extended by design.
We run regulated frontlines ourselves, so compliance is built into the platform, not bolted on. A compliance fix happens once and every agent inherits it.
Guardrails the model cannot override
Rules are checked by deterministic code outside the model. A persuasive prompt cannot talk its way past them.
A contact clock that fails closed
Contact rules are checked before every call or message. If a setting is missing, the contact is blocked, not waved through.
Consent kept as evidence
Consent is recorded with the action it allowed, so you can show why a contact was made, not only that it was.
No unconfirmed figures
An agent does not quote a balance, fee or settlement to a customer until the figure is confirmed from the system of record.
A person decides big actions
Routine actions never wait. Anything consequential pauses for a named person, the facts are checked again, then it acts.
Every step on a hash-chained ledger
Each action is written to the record before it runs. Change one entry and the chain no longer verifies.
See the audit ledger
Rules for the US, Canada and India are encoded as checks, and every agent must pass its evals before it goes live. See the controls we have mapped.
How to check us
Do not take our word for it.
Every promise on this page points at something you can open and test.
Verify a chain yourself
Tamper with a sample ledger in your browser and watch verification stop at the exact entry you changed.
Open the ledger sandboxRead the controls
Browse each framework, the controls we have mapped and the evidence behind each one.
Explore the frameworksSee how we protect data
Security in plain words: where your data lives and which models touch it.
Visit the trust centerChoose where it runs
Managed cloud, in your region or in your own cloud account. Pick what your regulator expects.
See deployment options
See both promises on a live conversation.
We will show a governed conversation end to end: the rulebook, the checks, the person who approves and the ledger entry that proves it.