Operations / October 1, 2026

AI utilization

Cost, tokens, reliability, and local compute in one view.

Representative workspace. Import your first usage file to replace this sample data automatically.

Estimated spend

$8.58

30d across all providers

Total tokens

2.43M

1.93M input · 497.04K output

Successful runs

95.8% success rate

Average latency

13.3s

96 normalized records

Spend over time

Daily cost by provider

OpenAIAnthropicGoogleOllamaOther

Monthly control

Budget

$0.00of $100.00

0% used30 days left

Projected month-end

$0.00

$100.00 under

Projection uses your current month-to-date run rate.

Model economics

Highest spend

gemini-flashGoogle

$2.9534.3%

claude-sonnetAnthropic

$2.6731.1%

gpt-5.4OpenAI

$2.1925.5%

gpt-5.4-miniOpenAI

$0.768.8%

qwen2.5:7bOllama

$0.020.2%

Data quality

Attention

5 items

5 events lack workflow labelsWorkflow labels improve cost attribution and client reporting.

One question · checked evidence · separate review

Diagnostic Team

Brief sets the scope. Meter checks costs. A researcher extracts evidence, a separate review pass challenges it, and Brief prepares the decision packet.

One workflow, up to three providers/models. Three model calls maximum; 1,800 output tokens per call maximum. No automatic retries or external actions. Review uses a fresh context but may use the same model. It is not a human reviewer.

One workflowCost CSV — provider,cost_usd,tokensApproved evidence notes — facts, source context, quality results and known gaps

Use synthetic or explicitly authorized sanitized inputs. This version analyzes supplied materials; it does not browse the web, read inboxes or retrieve customer records.

Execution path

Hosted Sites cannot reach Ollama on your laptop without the worker. Model/API charges may apply behind Hermes; dollar cost is unavailable and no hard dollar cap is claimed. Your server must explicitly enable this path after reviewing its model and billing configuration.

I approve passing these inputs to the selected runtime for up to three model calls.

Model-team execution is not enabled on this server. The input checker and downloadable baseline are available; runtime activation is still pending.

Keep this page open during a run. The assembled packet is not saved automatically; individual authenticated stage receipts use the existing run ledger. Download the packet before leaving.

Start with a deliverable

Revenue workflows

Prepare a report you can review, demonstrate and use in a scoped client engagement.

Prepare the measured cost baseline for a $750 founding diagnostic. The full engagement also needs quality testing and a findings walkthrough.

Open CSV file

CSV with headers: provider, cost_usd, tokens

These workflows use calculation and matching rules in your browser. Inputs are not uploaded or saved. Download the report before leaving.

$750 diagnostic delivery packet

One workflow, one reporting period, up to 100 usage rows and 10–50 human-scored quality cases. Target delivery: five business days after complete inputs and scope acceptance. Includes a findings walkthrough and one factual correction round.

Use the usage CSV above. Add evidence from a real review below; a model cannot approve its own findings. Inputs stay in this browser and clear when you leave.

Customer aliasWorkflow in scopeReporting periodHuman reviewerLimitations and unresolved questionsHuman-scored cases: case_id,result,critical_failure,review_minutes,evidence

Business connections

Check connection status to see which accounts still need setup.

Open Composio setup

Offer, guides and downloads →

Local GPU + cloud review workspace

Document workbench

Extract invoice fields and find relevant passages. Review the evidence and price the delivery work.

Run approved text on your computer. On the hosted site, open a local result for review; imported content stays in browser memory. Export before leaving. Search scores rank similarity, not correctness.

Prepare a local batch

Start the supplied local GPU launcher, then open this section there. Use 1–10 documents, each at most 1,800 characters, and 1–5 questions. Text only; PDFs and images need separate approved text extraction. Changing synthetic to false does not grant data-processing permission.

Approved batch JSON

Open local result

Quote and delivery economics

Editable planning assumptions, not measured profit. Include review, rework, sales time, cloud fees, equipment allocation and other costs. GPU electricity is only one small part of delivery cost.

Service price ($)Total owner time (minutes)Owner time value ($/hour)Whole-computer power assumption (watts)Electricity ($/kWh)Cloud costs allocated to delivery ($)Equipment, fees and other costs ($)

Estimated delivery cost: $275.00 · Contribution after valued owner time: $475.00 · Margin: 63.3%

GPU-run electricity estimate: $0.00000 — load a result to include run duration. Hosting subscriptions and standby energy must be allocated separately. No cloud-versus-local savings claim is made without equivalent quality and workload evidence.

Agent workspace

Command center

People, agents, tasks, approvals, channels, and computers in one customer-isolated workspace.

10 registered agents

0 verified active runs

0 approvals

0 connected runtimes

Give a specialist a jobDraft-only · run evidence persisted in Supabase

AgentExecution providerApproved business context

Start here

Choose the next revenue-producing action

A one-page decision brief with three ranked actions, a stop list, and a handoff.

Bring these inputs

Current offer
Evidence of buyer interest
Available time and budget

What good looks like

Rank by time to buyer evidence, effort, and cost—not excitement.
No invented commitments, owners, dates, or revenue.
Choose one measurable next milestone and a stop condition.

Examples need no customer data. Clear the task field to load one. Connections are not implied; these runs use only supplied material and return drafts.

Task or approved source material

No email, CRM, ticket, calendar, payment, or phone action can execute in this pilot.

Choose a job, load a synthetic example, and run a connected provider. Each specialist now has a defined deliverable and review checklist. Revised instructions are not a guarantee of output quality; use guided workflows above for checked calculations and reconciliation.

Agent rosterChoose a specialist

MeterBuild a defensible AI cost diagnostic

not connected

BriefChoose the next revenue-producing action

not connected

ScoutTurn one warm introduction into a discovery brief

not connected

RelayPrepare a support answer a human can trust

not connected

SignalTurn verified work into a buyer-ready proof pack

not connected

IntakeMake a messy document reviewable

not connected

PulseMake the weekly customer call worth attending

not connected

FoundryScope a custom agent before anyone builds it

not connected

CarePrepPrepare an administrative care-call packet

not connected

FlowReconcile conflicting records without losing evidence

not connected

Task boardBounded runs

1 · VALIDATEChoose one buyer problemBrief → Scout · owner reviews fit and outreach

2 · DELIVERProduce evidence, not promisesMeter or Intake → Flow · reviewed pilot artifact

3 · RETAINReview results with the customerPulse → Signal · only verified, approved proof

Approval queueHuman control

Approval queue is clear

Services and planned integrationsCustomer scoped

Hermes package readyGrok Bot package readyOllama not connectedSlack optionalTelegram optionalAgentMail optionalAgentPhone optionalRealtime voice opt-inGoogle Messages RCS / SMSOrgo VM optionalComposio setup requiredMeridian AI private-markets MCPPostgreSQL schema readySupabase Auth not connectedEncrypted secrets not connectedAudit store not connectedDurable queue not connectedExternal actions locked

One control plane, replaceable execution providersLocal PC, DGX Spark, Mac Studio, Hermes profiles, or isolated cloud computers can run behind the same policy and audit layer.

Local-safe demo

Product catalog

Forge Agents

Customizable role agents with secure defaults, approval gates, and portable deployment packages.

Policy engine included

ME$99–$499/mo

Cost & Operations Agent

Meter

Audits usage, finds waste, compares routing choices, and drafts an evidence-backed optimization brief.

Lower AI cost without reducing useful output

2 approval gates

SC$299–$1,500/mo

Sales Qualification Agent

Scout

Researches opted-in prospects, scores fit, drafts personalized follow-up, and prepares a CRM handoff.

Turn inbound interest into qualified, well-prepared conversations

3 approval gates

RE$399–$2,000/mo

Customer Support Agent

Relay

Classifies requests, retrieves approved knowledge, drafts grounded answers, and escalates risk or uncertainty.

Resolve routine questions safely and escalate the right cases

3 approval gates

BR$199–$999/mo

Chief of Staff Agent

Brief

Builds a daily brief, prepares meetings, extracts commitments, and tracks open decisions without sending on its own.

Convert scattered information into decisions and follow-through

3 approval gates

SI$149–$750/mo

Content Repurposing Agent

Signal

Transforms one source into channel-specific drafts while preserving voice, claims, and attribution.

Create more useful content from approved source material

3 approval gates

IN$500–$3,000/mo

Document Operations Agent

Intake

Extracts structured fields, validates required information, identifies exceptions, and prepares a review packet.

Turn unstructured documents into validated, routed work

3 approval gates

PUIncluded in managed plans

Customer Success & Retention Agent

Pulse

Prepares the weekly customer call, surfaces adoption and unresolved issues, tracks commitments, flags retention risk, and drafts the human-led follow-up.

Turn weekly customer conversations into measurable adoption, solved problems, and stronger retention

4 approval gates

FO$2,500–$15,000 setup

Agent Solutions Architect

Foundry

Guides discovery, selects or adapts a template, proposes tools and approvals, generates evaluation cases, and prepares a deployment package for human review.

InsuranceHealthcare administrationProfessional services+2

Turn an approved customer workflow into a scoped, testable agent package

5 approval gates

CAEnterprise pilot

Care Management Preparation Agent

CarePrep

Builds a cited pre-call brief from approved member, claims, care-plan, and program data; highlights gaps and questions; and routes the brief to a licensed human for review.

Health plansCare management organizationsProvider-sponsored plans+2

Help authorized care managers prepare for member conversations with less chart-review time

5 approval gates

FL$1,000–$7,500/mo

Data Automation Agent

Flow

Ingests documents and tables, maps fields to a customer schema, reconciles conflicts, and routes exceptions with full provenance.

Insurance operationsHealthcare administrationFinancial operations+3

Convert messy operational data into validated, reviewable workflows

4 approval gates

Secure runtime, not prompt-only guardrailsDefault-deny tools · human approval for external writes · per-run cost limits · prompt-injection checks · redacted audit events · kill switch

Policy tests

Normalized ledger

Recent activity

96 records

Time	Provider / model	Workflow	Tokens	Latency	Cost	Status
Yesterday 11:07 PM	Googlegemini-flash	Document extraction	18.59K	8.4s	$0.17	success
Yesterday 6:05 PM	OpenAIgpt-5.4	Unassigned	18.15K	16.9s	$0.10	failed
Yesterday 11:39 AM	Ollamaqwen2.5:7b	Local quality gate	12.99K	5.0s	$0.00	success
Yesterday 9:21 AM	Anthropicclaude-sonnet	Proposal critique	37.52K	11.2s	$0.12	success
Yesterday 3:14 AM	Googlegemini-flash	Document extraction	38.52K	5.1s	$0.14	success
Sep 29 8:07 PM	OpenAIgpt-5.4	Architecture review	9.85K	4.9s	$0.11	success
Sep 29 4:05 PM	OpenAIgpt-5.4-mini	Routing experiment	24.65K	19.6s	$0.05	success
Sep 29 2:47 PM	Anthropicclaude-sonnet	Proposal critique	38.66K	23.2s	$0.14	success
Sep 29 10:20 AM	OpenAIgpt-5.4	Architecture review	10.06K	3.9s	$0.06	success
Sep 29 3:33 AM	OpenAIgpt-5.4-mini	Routing experiment	30.71K	21.1s	$0.04	success
Sep 28 5:35 PM	Anthropicclaude-sonnet	Proposal critique	10.51K	23.4s	$0.05	success
Sep 28 5:33 PM	Ollamaqwen2.5:7b	Local quality gate	41.17K	13.0s	$0.00	success
Sep 28 12:15 AM	Googlegemini-flash	Document extraction	34.7K	10.0s	$0.20	success
Sep 27 5:42 PM	OpenAIgpt-5.4	Architecture review	7.91K	19.7s	$0.04	success
Sep 27 10:40 AM	OpenAIgpt-5.4-mini	Routing experiment	21.62K	23.0s	$0.05	success
Sep 27 7:48 AM	Ollamaqwen2.5:7b	Local quality gate	18.62K	3.9s	$0.00	success
Sep 27 3:25 AM	Ollamaqwen2.5:7b	Local quality gate	38.89K	11.0s	$0.00	success
Sep 27 1:20 AM	Googlegemini-flash	Document extraction	14.64K	17.7s	$0.15	success
Sep 26 9:10 PM	Anthropicclaude-sonnet	Proposal critique	36.59K	11.9s	$0.17	success
Sep 26 7:57 PM	OpenAIgpt-5.4-mini	Routing experiment	16.09K	10.0s	$0.05	success