Synthetic / read-only pilot console showing how JoshOps AI would govern AI workflows — model throttling, approval gates, connector readiness, evidence receipts, synthetic artifact review, and scoped pilot planning. Mock data only. No live connectors. Not certified.Home

JoshOps AI

JoshOps AI • Ready/pilot/throttle

ARES

online · policy

Sphinx

idle · scan

Enterprise Pilot Console

Model governance & throttle

Per-model token budgets, decision rate, and current hold rate.

Local JoshOps AI offline — sample data.

62%

reasoning

Claude Code

ceiling 100%

83%

drafting

JoshOps AI Mark

ceiling 90%

36%

policy

Mistral 7B Instruct

ceiling 100%

scan

Llama 3.1 8B

ceiling 100%

25%

reasoning

Claude Opus 4.1

ceiling 100%

39%

reasoning

GPT-5 Pro

ceiling 100%

42%

scan

Gemini 2.5 Pro

ceiling 100%

42%

drafting

Grok 4

ceiling 100%

67%

reasoning

DeepSeek V3

ceiling 100%

38%

reasoning

Llama 3.3 70B

ceiling 100%

16%

local

Whisper v3 + Piper

ceiling 100%

Decisions / min

243

across 4 agents

Spend / hr

$8.24

cloud models only

gCO₂ / hr

297

~ 7.13 kg/day

Egress posture

2 of 4 local

1 air-gapped · 2 cloud (policy-gated)

[ 02 :: LIVE ROUTE TRACE ]

Why this model · what A.R.E.S. allowed · what Sphinx verified · what receipt was written

JoshOps AI does not just call a model. It records why a model was allowed, why others were held, what authority was granted, and what receipt was written.

allowedheldblockedselected modelskipped / standbyintent · model governance check

Step detail

hold

A.R.E.S. policy gate

authority + risk

Read-only granted. Mutations require bounded work order.

authority_state

observation_only

next_safe_action

write_observation_receipt

blocked_actions

commitpushdeletepromote_canonupdate_profile_truth

[ ROUTE RECEIPT ]

route_20260620_0914 · authority observation_only

{
  "schema": "memex_route_trace_receipt_v0",
  "route_id": "route_20260620_0914",
  "intent": "model governance check",
  "authority_state": "observation_only",
  "ares_decision": "allow_observation_only",
  "selected_model": "mistral:latest",
  "held_models": [
    "claude_code"
  ],
  "skipped_models": [
    "memex_mark"
  ],
  "blocked_actions": [
    "commit",
    "push",
    "delete",
    "promote_canon",
    "update_profile_truth"
  ],
  "sphinx_verification": "passed",
  "next_safe_action": "write_observation_receipt",
  "mutation_allowed": false
}

[ MODEL ELECTION ]

why one was selected · why others were held, skipped, or on standby

Model	Role	Locality	Cost	Risk	Decision	Reason
Mistral 7B Instruct	local semantic witness	local	$0.00	low	selected	Enough for local meaning check; zero egress.
Claude Code (Sonnet 4.5)	patch worker	cloud	$1.42/1k	medium	held	Patch work requires bounded work order — not granted.
JoshOps AI Mark (gpt-4o-mini)	drafting voice	cloud	$0.18/1k	medium	skipped	No drafting needed for this route.
Llama 3.1 8B	air-gapped scan	air-gapped	$0.00	low	standby	Available; not invoked outside scan window.

[ 02 :: AGENTIC FLOW ]

Decision path · intent → policy → model → verify → receipt

allowholdblock

[ 03 :: AGENT ROSTER ]

model · locality · cost · carbon · 24h utilisation

Model · roleLocality$/1kgCO₂/hr24h utilisationStatus

Claude Code (Sonnet 4.5)
Anthropic · cloud · Patch worker · code edits
Cloud$1.42/1k · 38gCO₂/hr
Cloud
$1.42
38
62% now · hold 12%
ok
JoshOps AI Mark (gpt-4o-mini)
OpenAI · cloud · Drafting · summaries · receipts
Cloud$0.18/1k · 22gCO₂/hr
Cloud
$0.18
22
83% now · hold 4%
throttling
Mistral 7B Instruct
Ollama · local · ARES policy check · on-device
Local · on-device$0.00/1k · 12gCO₂/hr
Local · on-device
$0.00
12
36% now · hold 31%
ok
Llama 3.1 8B
Ollama · local · Sphinx scan · folder + leak canary
Air-gapped$0.00/1k · 8gCO₂/hr
Air-gapped
$0.00
8
0% now · hold 0%
paused
Claude Opus 4.1
Anthropic · cloud · Long-horizon planner · constitution drafts
Cloud$3.85/1k · 64gCO₂/hr
Cloud
$3.85
64
25% now · hold 22%
ok
GPT-5 Pro
OpenAI · cloud · Cross-tenant arbiter · escalation
Cloud$2.10/1k · 48gCO₂/hr
Cloud
$2.10
48
39% now · hold 18%
ok
Gemini 2.5 Pro
Google · cloud · Long-context audit · 2M token corpus sweep
Cloud$0.85/1k · 31gCO₂/hr
Cloud
$0.85
31
42% now · hold 9%
ok
Grok 4
xAI · cloud · Open-web drafter · sandboxed
Cloud$0.92/1k · 27gCO₂/hr
Cloud
$0.92
27
42% now · hold 14%
ok
DeepSeek V3
DeepSeek · cloud · Cost-tier reasoner · bulk triage
Cloud$0.14/1k · 19gCO₂/hr
Cloud
$0.14
19
67% now · hold 8%
ok
Llama 3.3 70B
Ollama · local · Sovereign reasoner · air-gap tenants
Air-gapped$0.00/1k · 22gCO₂/hr
Air-gapped
$0.00
22
38% now · hold 19%
ok
Whisper v3 + Piper
Local · on-device · Voice membrane · speech ↔ text
Local · on-device$0.00/1k · 6gCO₂/hr
Local · on-device
$0.00
6
16% now · hold 2%
ok

[ 05 :: FAILOVER & QUORUM ]

What happens when something disagrees, drops, or doubts.

IfClaude Code unavailableGPT-4o → o4-mini
warm standby · <800ms cutover
IfARES policy disagrees with Sphinx scanHold · human approval
approver: tenant ops lead
IfLocal model confidence < 0.62Escalate to reasoning tier
logged + replayed nightly

[ 04 :: LIVE DECISION STREAM ]

streaming · synthetic

--:--:--allowMistral 7B · folder-scan/Q3-vendors1,240 tok
--:--:--allowClaude Code · builder/patch · receipts.ts4,820 tok
--:--:--holdJoshOps AI Mark · draft/email · acme-renewal612 tok
--:--:--allowLlama 3.1 8B · leak-canary/exports980 tok
--:--:--blockMistral 7B · connector/salesforce · raw PII—
--:--:--allowClaude Code · builder/patch · gates.tsx2,310 tok
--:--:--allowJoshOps AI Mark · summary/sop-drift · ops-121,820 tok
--:--:--holdMistral 7B · approval/tenant-budget +18%240 tok

Local · Mock · Synthetic

⌘K next action · ⌘/ talk