Built for permission changes, not ordinary notes.
The ledger records what changed about the agent's authority. It separates harmless preference memory from changes that alter whether an agent can proceed, batch, ask, or block future work.
Consent ledger software for personal AI agents.
Track every meaningful change to an agent's permission memory: the old boundary, new boundary, evidence, approval route, reviewer, and recovery path before autonomy expands.
Niche landing page
A consent ledger gives personal agents a control surface for learned permission.
Personal agents can learn from corrections, but learned memory should not quietly rewrite consent. Consent ledger software turns sensitive memory changes into structured entries that can be reviewed, approved, narrowed, expired, and audited over time.
Before and after
Every entry stores the old boundary and proposed new boundary in plain language.
Evidence attached
The correction, rejected draft, approval message, or replay travels with the rule change.
Review route
High-risk entries go to text approval; low-risk entries can wait for a digest.
Super as the approval surface
Super can carry urgent consent-ledger decisions to the operator. The text message AI assistant use case is a natural route for approve, narrow, or reject actions.
Proof for delegated browser work
For browser tasks, ledger entries should include replayable evidence. The computer-use cache and AI website-building workflow show how proof supports safer autonomy.
Workflow
What happens when a correction might change consent.
The product should make a sensitive memory update feel less like a hidden prompt edit and more like a clear operational change request.
Detect a boundary-changing correction.
Flag edits that affect privacy, external communication, browser action, spending, or approval thresholds.
Create the ledger entry.
Write the old rule, new rule, source event, scope, consequence, and recommended review lane.
Route by consequence.
Immediate text approval for sensitive changes, batch review for moderate changes, digest for low-risk entries.
Store the receipt.
Once accepted, save reviewer, timestamp, expiration, and rollback instructions so the rule remains inspectable.
Decision lanes
Ledger entries should land in a clear lane.
Proceed
Action can run with a receipt.
Batch
Review later without interrupting.
Ask
Interrupt by text before action.
Block
Prevent the future action class.
Checklist
Minimum viable consent ledger fields.
Old boundary What the agent believed before.
New boundary What the update would change.
Source event The correction or evidence behind it.
Scope Person, project, channel, task, or date range.
Escalation lane Proceed, batch, ask, or block.
Recovery Expiration and rollback path.
FAQ
Questions operators ask about consent ledger software.
Is this different from a normal audit trail?
Yes. An audit trail records agent actions. A consent ledger records changes to the agent's future permission boundary.
Should all memories go into the ledger?
No. The ledger should focus on memory changes that affect approval thresholds, private context, communication, browser work, spending, or reputation.
Where should ledger approvals happen?
Urgent and sensitive approvals should happen by text. Lower-risk entries can be reviewed in a dashboard or digest.
Give agent memory a ledger.
Consent ledger software makes learned autonomy reviewable: every sensitive rule change gets evidence, scope, approval, and recovery before it shapes future work.