Usage value, not an invoice
Your AI factory is visible now.
Loading model, agent, tool, and cost evidence from this machine.
Recorded usage value
$0.00
0 AI credits
AI operations / command center
Command center
Model routing
Credit distribution
Maturity
Assessing...
Orchestration split
Who consumed the credits?
Human loop
Attention and steering
Session ledger
Where work and cost accumulated
Session cost distribution (credits)
Model calls vs. credits (bubble size = tool calls)
Model economics
Calls, tokens, and credit intensity
| Model | Calls | Input | Output | Cache read | Credits | Share | Avg / call |
|---|
Token composition
Input, output, cache, and reasoning mix
Cache efficiency
Reuse ratio
Credit intensity
Calls vs. average credits per call (bubble size = total credits)
Orchestration
Agent routing and model mix
Bar length is each agent type's specialist credit total. Segments show which models it reported using; an exact per-model credit split per agent is not recorded, so segments are shown as an even share and labelled accordingly.
Reliability
Success vs. failure
Aggregated across every completed tool call in this window.
Frequency vs. footprint
Calls vs. estimated context footprint
Bubble color signals recorded success rate: accent = healthy, amber = watch, red = degraded. Size reflects failure count.
Tool inventory
Calls are frequency. Footprint is estimated context.
The call-share bar is not a token bar. Estimated I/O uses UTF-8 payload characters divided by four and is never presented as billable usage.
| Tool | Calls | Call share | Success | Failures | P95 | Est. input | Est. result | Top model |
|---|
Evidence-based orchestration review
Strong execution. Missing factory guardrails.
The audit separates recorded evidence from recommendations and never treats token-footprint estimates as billing facts.
Current stageL3Orchestrated
Maturity
Domain radar
Input audit
Human vs. delegated vs. system inputs
Input audit
How autonomy was initiated and steered
Interpretation
Target operating model
From orchestrated swarm to controlled dark factory.
Autonomy scales only when intake, budgets, evidence, independent assurance, and human escalation are first-class control-plane services.
Release policy
Six non-negotiable gates
Reusable hook
Human attention manager
One service owns notifications, prompts, timeouts, deduplication, and escalation across every agent team.
Connector capability matrix
Live, imported, estimated, or unavailable. Never ambiguous.
Credentials stay on the loopback service. Consumer products without a supported usage API remain import-only instead of being scraped through unstable private stores.
Live
Ready
Configure
Import only
Local runtime and evidence
Operate the loopback service with evidence, not guesswork.
Events are polled from a bounded local ledger. They contain status and aggregate evidence only—never prompts, code, tool payloads, paths, or credentials.
Runtime
Service status
Advisor
Evidence-based next actions
Deterministic local analysis only. It does not inspect prompt content or call an LLM.
Live event ledger
Polled local events
Session mapping
Explicit local work-item link
This saves only a local mapping. Jira, Azure DevOps, and GitHub network integrations are future architecture, not active calls.
Saved mapping ledger