Skip to main content
Calling external parsing APIs incurs costs. AllCodex includes token and budget observability dashboards to monitor spending and enforce daily limits.

​
The observability dashboard

The Observability and Settings pages in the Portal display:
  • Daily burn rate: A chart showing your API cost over the last 30 days.
  • Task breakdown: Cost split by operation type (such as Brain Dumps, Consistency Checks, and Relationship Suggestions).
  • Model breakdown: The cost split by model (e.g. Grok-4 vs Kimi-k2.5).
  • Token counts: Split by prompt (input) and completion (output) tokens.

​
Budgets and hard limits

To protect self-hosters from runaway loops or excessive usage, AllCodex supports user-scoped budgets:
  1. Daily warning threshold: Triggers a warning badge in the Portal navigation bar when daily spending crosses this amount (e.g. $1.00/day).
  2. Hard daily budget: Blocks subsequent AI calls for the remainder of the day once reached (e.g. $2.00/day). AllKnower returns a 403 Forbidden error.
  3. Defaults: Standard budget defaults to a conservative $0.50/day.

​
Database tracking schema

Three tables track observability metrics in AllKnower’s PostgreSQL database:

​
llm_call_log

Logs every API request sent to OpenRouter.
  • Fields: id, userId, taskType (e.g. brain_dump), model, promptTokens, completionTokens, cost, timestamp.

​
model_pricing

Maintains a cache of OpenRouter model prices (input cost per 1M tokens, output cost per 1M tokens) to calculate transaction costs locally.
  • AllKnower pulls fresh pricing from OpenRouter via POST /usage/refresh-pricing during non-production updates.

​
user_budgets

Stores the warning and limit thresholds for each user.
  • Fields: id, userId, dailyWarningLimit, dailyHardLimit.