Platform overview
Live telemetry across accounts, inference runtime, token reservations, and ledger settlements.
Service boundaries and health
- Application API
- Online
- Inference runtime
- Checking...
- Checkpoint path
- -
- Repository mode
- Portable repository interface
Configured plan catalog
| Plan | Monthly price | Monthly quota | Max output | Rate limit |
|---|---|---|---|---|
| Loading configured plans... | ||||
User management and account governance
Inspect registered accounts, update account state transitions, or assign administrative roles with mandatory audit logging.
| User ID | Email and name | Plan | State | Roles | Actions |
|---|---|---|---|---|---|
| Sign in with an admin account to load users. | |||||
Update user state or roles
Subscriptions and quota adjustments
Perform server-verified plan changes or compensating ledger adjustments. Restricted to finance_admin and super_admin roles.
Manual subscription or ledger adjustment
Model and inference runtime
Inspect Pearl Chat v2 model parameters, ChatML guardrails, and trigger safe checkpoint reloads.
Active model configuration
- Model family
- Pearl Chat v2
- Architecture
- RoPE, GQA, SwiGLU, Pre-RMSNorm
- Total context ceiling
- 2,048 tokens
- Tokenizer
- 32,768-vocab Luganda BPE
- Supported task modes
- dialog, continuation, translate
Server-enforced generation guardrails
- Temperature bounds: 0.00 to 1.20 validated server-side.
- Top-p bounds: 0.10 to 1.00 validated server-side.
- Top-k bounds: 1 to 100 validated server-side.
- Repetition penalty: 1.00 to 1.30 validated server-side.
- Deterministic context trimming: Oldest turns are removed when input tokens plus max_new_tokens exceed 2,048 tokens.
Immutable audit logs
Every administrative action, account state transition, and manual subscription change is recorded with actor and reason metadata.
| Timestamp (UTC) | Action | Actor | Target | Reason and metadata |
|---|---|---|---|---|
| No audit events loaded yet. | ||||