Skip to main content
When this applies: you’re an admin (owner / admin role) on a ByteSpike org, and you need to grant employees model access, monitor usage and balance, or handle abuse or model-availability incidents. This is the companion piece to Employee onboarding — that one tells employees “how to use it”, this one tells admins “how to run it”.

1. Three surfaces at a glance

ByteSpike currently lives on 2 domains + 1 sub-path:
Domain / pathPrimary audienceWhen you (admin) use it
llm.bytespike.ai/adminPlatform backendMain entry point right now — user management, available models, model groups, audit logs
console.bytespike.aiUser-tier viewView your own keys / usage / billing, enterprise dashboard, hand out keys to employees
Eventually admin.bytespike.ai will be split out as its own subdomain (cleanly separate from the user-facing console). Right now the admin backend still lives under the llm.bytespike.ai/admin path, sharing the hostname with the inference gateway. This document uniformly refers to llm.bytespike.ai/admin — once migration is done we’ll update.
90% of day-to-day operations happen at llm.bytespike.ai/admin. The console is for seeing your own user-side view (top-up, calling the API, looking at your own row in the audit log).

2. Grant employees model permissions (core operations action)

When a new employee joins, 0 models are usable by default. You need to grant them.
Public groups don’t need to be written into the user_allowed_groups table — the backend uses User.CanBindGroup(_, is_exclusive=false)=true and waves them through. So even if the admin UI shows an employee’s permissions as “unassigned”, every public group is still callable. Exclusive groups are the ones that need explicit grant rows in the table.

3. Usage monitoring + balance alerts

1

Daily usage dashboard

llm.bytespike.ai/admin → Usage → group by day / model / api_key / user across multiple dimensions. CSV export available.
2

Overview (dashboard)

llm.bytespike.ai/admin → Overview → today’s total usage, tier breakdown, active users, balance trend.
3

Low-balance alerts

llm.bytespike.ai/admin → Settings → balance_notify → set a threshold (e.g. $50) + recipient email. When the org wallet drops below the threshold, cron sends an email + writes a system.balance.notify.dispatched notification into the console notification center.Mechanism details at /me/notifications.

4. Model permissions (platform layer)

Different from per-employee permissions — this governs “which models the ByteSpike platform itself can call”.

Model Groups

llm.bytespike.ai/admin → Model Groups — add models to a group, set multiplier (e.g. put a model in a 0.8x discount group), set the exclusive flag (whitelist-only access).

Available models registry

llm.bytespike.ai/admin → Available Models — when a new model is added to the catalog, register it here + associate the model group.

API Keys (your own)

console.bytespike.ai/keys — your own admin test keys. For production keys, employees generate them inside DOSIA, or admin batch-issues them at console.bytespike.ai/org/keys.

Audit log

llm.bytespike.ai/admin → Audit log — 25+ action types covering every write operation across admin / org / user layers. Each entry carries actor + IP + UA.

5. Top-up and billing

1

Top up the org wallet

console.bytespike.ai/billing → pick Stripe (visa / mastercard / amex) or Alipay → enter amount → done. The webhook receives system.payment.received and balance lands instantly.
2

Auto-recharge policy (optional)

On the billing page, Auto-recharge section — set a trigger threshold (e.g. balance below 50)+anautochargeamountpertrigger(e.g.50) + an auto-charge amount per trigger (e.g. 200). Charges happen automatically via the Stripe saved payment method.
3

Billing history / export

console.bytespike.ai/billing → Transactions → complete ledger (topup / refund / admin_credit / admin_debit / subscription_renew). Each row has CSV export, meets finance-team requirements.API endpoint at /me/billing/transactions.

6. Emergency response SOP

The full operations troubleshooting (balance crash / model-availability incidents / employee abuse / mistaken revocation + handling employee feedback) lives in troubleshooting · Admin view tab. That page also collects every employee blocker from Employee onboarding, organized by role + scenario in AccordionGroups.

What’s next

For operations questions or corner cases that need OP intervention: ping the ops team / W1.