Key facts
| Event capture | One structured event per model call, tool action and approval |
| Platform events | Authentication, key lifecycle and admin changes exportable from Plugsky |
| Tenant attribution | Per-tenant keys and scoped service identities |
| Delivery | In-app activity view, REST export and webhooks you build on the event stream |
| Retention tiers | Short default windows with longer retention on higher plans |
| Content control | Metadata-first; content capture only as an explicit, consented option |
| Deployment | Cloud, VPC and on-prem options for enterprise tenants |
| Status | Audit export is live; assistants and responses endpoints are coming soon |
TL;DR
- Capture once, present many ways: in-app, API, webhook and export.
- Attribute every event to a tenant, user and feature.
- Metadata-first; make content capture an explicit enterprise option.
- Retention tiers are a natural upgrade lever.
- Enterprise buyers ask for this during security review — having it shortens sales cycles.
How it works, step by step
- Define the event schema for tenant-visible activity: timestamp, actor, action, resource, model, outcome.
- Emit events at the service layer where tenant context already exists, and add platform events from audit export.
- Store events per tenant with a retention tier matched to the customer plan.
- Build an in-app activity view for administrators, with filters by user, feature and time.
- Expose an API and webhooks so enterprise tenants can feed their own SIEM or compliance tooling.
- Add CSV or JSON export for audit requests, and document exactly what is and is not recorded.
- Make content capture an opt-in enterprise capability with short retention and access controls.
Try it yourself
Open the AI API key security checklist →
From internal logging to product surface
Internal logs answer your questions; product logs answer your customer's. The same underlying events can serve both if the schema is good: timestamp, actor, action, resource, model and outcome, attributed to a tenant and a feature. Capture once at the service layer, then present through an in-app activity view, an API, webhooks and file export.
Plugsky contributes platform-side events — authentication, key lifecycle and administrative changes — through audit export. Merge those with application events so a customer can see not only what the AI did but also who changed the keys and settings that govern it.
Design choices that sell
Security reviews ask predictable questions: what do you record, where is it stored, how long is it kept, and can we get it out? Answering those in the product rather than in a spreadsheet is a competitive advantage, particularly for enterprise deals in regulated sectors.
- Retention tiers: short windows by default, longer retention on higher plans — a clean upgrade lever.
- Export: API, webhooks and file export, with a documented schema.
- Content: metadata by default; content capture as an explicit, consented enterprise option.
Privacy and operational discipline
Tenant-visible activity is a privacy surface. A workspace administrator should see what their organisation did, not read another user's private prompts. Keep content out of the default view, apply role-based access, and be precise in documentation about what metadata reveals — model, feature, timing and outcome are usually enough.
Operationally, treat the event pipeline like any other production data path: schema versioning, backfill strategy, retention enforcement and monitoring. When enterprise customers rely on your feed for their own audit evidence, availability and correctness of that feed become part of your SLA, not a nice-to-have.
Honest comparison
| SaaS decision | Plugsky plus your app | Typical model API | In-house build |
|---|---|---|---|
| Platform events | Auth, key and admin export included | Limited | You build them |
| Per-call metadata | Model, tokens, latency, request IDs | Partial | Your instrumentation |
| Tenant attribution | Per-tenant keys and scoped identities | Shared key | Your design |
| Delivery | Your API, webhooks and UI on the event stream | Not applicable | Your entire stack |
| Retention tiers | Your policy, your plans | Not applicable | Your storage |
| Enterprise readiness | Shortens security review | Not addressed | Months of work |
Frequently asked questions
Should tenant administrators see prompt content?
By default, no. Show metadata: who acted, which feature, which model, when and with what outcome. Content visibility should be a deliberate enterprise option with consent and retention controls.
What events should be tenant-visible?
AI actions, tool calls, approvals and overrides, plus key and administrative changes. Those cover the questions security teams and auditors ask most often.
Do we need webhooks if we have an API?
Both help. APIs suit dashboards and batch export; webhooks let customers stream activity into their own SIEM in near real time, which large tenants prefer.
How do we store events per tenant?
Store with a tenant identifier and enforce retention by plan tier. Partitioning by tenant keeps queries fast and offboarding clean.
Is this a compliance feature or a product feature?
Both. It satisfies security review requirements and gives administrators genuine operational value — seeing what automation did on their behalf.
How do we capture events from Plugsky?
Platform events come from audit export; application events you emit at your service layer where tenant context exists. Merge both in your event store.
How long should we keep events?
Set a short default for all tenants and offer longer retention on higher plans, aligned with your privacy policy and documented in customer-facing terms.