The audit trail is a pane: governance you can watch in real time

Jul 19, 20265 min readBy the Rysh team

Admin-console audit logs are past tense. A pane is present tense. Governance works when the trail lives where the work happens.

Every enterprise AI product has an audit trail. Ask for a demo and you'll get the same screen: an admin console, a filterable table, timestamps, actor columns, an export-to-CSV button. Governance, allegedly.

Here's the problem with that screen: it's past tense. By the time an event reaches the audit table, the thing already happened. The console is where you go to reconstruct the incident, attribute the blame, write the postmortem. It's a forensics tool wearing a governance costume.

Building rysh, an agentic terminal multiplexer, I've been betting on a different shape: the audit trail isn't a separate surface you visit after the fact. The audit trail is the pane — the same place the work happens, while it happens.

Present tense vs past tense

Watch what an agent run looks like in a rysh pane:

> migrate the user table to add soft-deletes

[tool] file_read migrations/0042_users.sql
[tool] file_write migrations/0043_users_soft_delete.sql
       + ALTER TABLE users ADD COLUMN deleted_at TIMESTAMPTZ;
       + CREATE INDEX idx_users_deleted_at ON users (deleted_at);
       [preview only — awaiting approval y/n]
y
[tool] bash: migrate up 1        [not on allowlist — approve? y/n]
y
       → applied 0043_users_soft_delete

Every element a compliance officer wants in an audit log is here — the action, the artifact, the human decision, the sequence. But notice the tense. The approval isn't recorded; it's happening. The diff isn't an entry describing a change; it's the change itself, pre-execution, waiting. The trail and the control point are the same object.

That's the actual definition of governance, as opposed to auditing: intervention has to be possible at the moment the record is created. An admin console can't offer that. A pane can't avoid it.

The trail persists — and it's queryable

"Live" doesn't mean "ephemeral." Pane conversations in rysh persist across detach and restart (state is stored locally via embedded NATS JetStream — your disk, your retention). Monday you can scroll Friday's session. The trail is the transcript; the transcript survives.

Better: it's queryable from inside the system. rysh exposes a session_history tool, so retrieving what happened in any pane is available to humans and agents alike. A reviewer agent can read a worker agent's transcript before signing off on its output. Your audit trail just became an input to your governance process, not only an output of it.

And when review needs more than scrollback, the pane itself is shareable: ##share pane in view mode lets a teammate watch a live session from their own terminal — read-only, real time. "Shadow the new agent workflow" stops being a metaphor. (Treat a shared pane like a screen share — what's visible is what you're showing.)

Same idea, outward-facing: draft-and-confirm

The pane-as-control-point pattern extends beyond your terminal. rysh humanoids — agents wired to external channels — carry per-channel governance: governance ai for autonomous replies, governance human for draft-and-confirm, where every outbound message the agent composes appears for a human to approve before it's sent (live on the email channel today).

Same tense argument, higher stakes: an audit log that shows what your support agent sent to a customer is forensics. A draft sitting in front of a human before it sends is governance. Start channels in human mode, watch the drafts, earn the confidence to flip specific channels to autonomous — with the switch, either way, in your hands.

"But a real audit system has—"

Yes. Let's be honest about what a pane is not:

Not tamper-evident compliance tooling. rysh's trail is the persisted transcript on infrastructure you control. If your regulator requires cryptographically signed, immutable audit records, that's a dedicated system — run it alongside; nothing here conflicts.
Not automatic org-wide aggregation. The pane model is deliberately local-first: trails live with sessions, on your machines, not auto-shipped to a vendor cloud. That's an ownership feature with a real trade-off — cross-team rollups are yours to build on top, not built in.
Not a substitute for reading. A visible approval can still be waved through. The design keeps prompts rare enough to deserve attention (the gate design), but no tool fixes a reviewer who won't review.

The claim isn't that a pane replaces enterprise audit infrastructure. The claim is about where governance actually happens: at the moment of action, in front of the person doing the work. A control that lives there gets used. A console that lives elsewhere gets visited quarterly.

The everyday-ness is the point

Here's the observation that convinced me. Ask a team using a black-box agent platform "what did your agents do yesterday?" and someone opens a dashboard, filters, squints. Ask a rysh user and they scroll up.

Governance that requires a special trip decays. Governance that's ambient — the trail scrolling past as you work, the approval interrupting exactly when it should, the teammate's pane one ##share away — compounds. People govern what's in front of them. So put the trail in front of them.

The audit trail is a pane. It was always supposed to be.

rysh is in private beta — self-hostable, your keys, your data, built on Claude. If your AI governance story currently lives in a console nobody opens, come be a design partner: free access, direct founder line, and your compliance requirements become roadmap. Apply here.

Try Rysh

Every pane is a shell and an AI agent — install takes one command.

Get started free →