Docket
Azure FinOps built for the speed you actually need it. Docket combines daily FOCUS billing exports for invoice-accurate reconciliation, near-realtime Azure OpenAI diagnostic telemetry for instant spend triage, and AI-generated optimization recommendations — with a PowerShell resource management suite that can act on what it finds.
curl "https://docket.celladoresystems.com/api/v1/spend?workspace=ws_prod&period=current_month" -H "Authorization: Bearer TOKEN" - Three latency tiers
- ~1 min realtime → 1–5 min near-realtime → 4–24h FOCUS billing
- gpt-4.1-mini AI Advisor
- weekly ranked optimization recommendations from accumulated cost data
- Zero manual infrastructure
- provisioned end-to-end with Terraform; no manually-created Azure resources
How Docket Connects to the Stack
Celladore is engineered as a unified, composable toolchain where every tool feeds upstream contexts and downstream verification loops without friction.
→ Sluice
Sluice emits AOAI diagnostic events that stream through Event Hub into Docket's ADX aoai_logs_raw table — priced per-request in near-realtime.
→ Retort
Correlates Retort agent run IDs with exact dollar amounts and token metrics via sluice OTel data for complete auditability.
→ Deck
Deck reads aggregated spend summaries via Docket's MCP server at /mcp (JSON-RPC over HTTP) for Dashboard cost tiles.
Technical Matrix
- Runtime & Engine:
- Python 3.12+ / FastAPI / KQL / PowerShell 7.0+
- Deployment Target:
- Azure Container Apps + Azure Data Explorer (pvc-docket-kusto)
- Communication Protocol:
- OTLP :4317 / Prometheus / Event Hub Streaming / FOCUS CSV
- Licensing Tier:
- Proprietary (Private — Internal Infrastructure)
Invented from Real Scars
Lack of agent-level cost attribution meant runaway recursive loops were discovered weeks later on aggregate monthly cloud invoices — with no way to trace which agent, task, or repo caused the spike.
Runaway agent loops bankrupting cloud budgets and unallocated venture LLM spend.
Architectural Features & Guarantees
Modular design built for high velocity, deterministic reliability, and strict enterprise isolation.
Three-Tier Cost Architecture
Tier 1 (~1 min) Azure Monitor metrics in Grafana. Tier 2 (1–5 min) AOAI diagnostic logs via Event Hub for per-request cost estimates. Tier 3 (4–24h) FOCUS-format exports for invoice-accurate billing reconciliation.
AI Cost Advisor
A gpt-4.1-mini analysis job runs weekly against all accumulated ADX cost data and produces ranked, actionable optimization suggestions stored in the cost_recommendations table.
PowerShell Resource Management Suite
Eight production-grade scripts handling inventory, deletion, lifecycle, suspension, scaling, tagging, health audit, and workflow orchestration — with IaC detection, cascade depth limits, and DryRun mode.
KQL Anomaly Detection
Built-in KQL functions: cost_top_resources(), aoai_cost_estimate(), aoai_anomaly_detect() — surface untagged resource burn and usage spikes before the monthly invoice lands.
Grafana Dashboards & Alerting
Five versioned dashboard JSONs covering all-resource cost overview, realtime AOAI cost triage, cost centre breakdown, and staleness/anomaly alerts.
Cost Centre API
FastAPI service tracking which teams and cost centres are onboarded. Backed by Azure Data Explorer in production, Table Storage as fallback for local dev.
Typed API Contract & Snippets
Live cURL, TypeScript, and Python invocation payloads for Docket.
Unified Developer Interface
Query real-time token spend rollups, budget consumption by agent, and cost alert thresholds.
import { DocketClient } from "@celladore/docket-sdk";
const docket = new DocketClient({
apiKey: process.env.DOCKET_API_KEY || "YOUR_DOCKET_TOKEN",
endpoint: "https://docket.celladoresystems.com",
});
const report = await docket.getSpendReport({
workspaceId: "ws_prod",
breakdown: "by_agent",
});
console.log(`Current LLM spend: $${report.totalCostUsd} (Budget: $${report.budgetCapUsd})`);Status Notice: Docket powers our internal FinOps and cloud cost operations across the Phoenix VC ecosystem. Architecture and telemetry schema are open for inspection. Contact us if you're building similar FinOps tooling for autonomous agent fleets. Get in touch with the founder .