Kettle Logic logo mark Kettle Logic
Start an assessment
Governed AI agents · MCP-native · B2B operations

Operations that run themselves
— and know when to ask.

Kettle Logic deploys governed AI agents that handle the operational busywork — reconciliation, exception triage, catalog QA — around the clock. They run unattended, and escalate the second a decision needs a human. Direct founder access · fixed-scope pilots.

Start a pilot → Watch the agents work →
24/7 autonomous Human in the loop MCP native Audit trailed
22
Live, documented APIs you can call
MCP
Model Context Protocol server, shipped
64✓
Verified deploy promises (Media Stack)
100/100
Lighthouse target on every page
What we build

Platform capabilities

One operating spine for B2B teams — from the data underneath to the agents, events, and dashboards on top.

  • 01 / command center

    Business command center

    Operator dashboards and the tooling to run the business — SLA status, queue aging, and the next best action.

  • 02 / data foundation

    Data foundations for AI

    Modeling, quality, and pipelines that turn messy operational data into trustworthy records — the groundwork AI depends on. Built on experience running data platforms at billions-to-trillions of records.

  • 03 / agents

    Automation and agents

    Agentic AI copilots for triage, enrichment, and routing — within governed, human-approved boundaries.

  • 04 / integration

    Integration and eventing

    API-first integration and event orchestration across ERP, POS, PIM/MDM, WMS, and partner channels.

  • 05 / observability

    Observability and reliability

    SLO-driven telemetry, synthetic checks, and runbooks that keep operations healthy and observable.

  • 06 / governance

    Security and governance

    Role-based access, policy controls, approvals, and change governance for multi-entity operations.

Your AI workforce · 50+ agents, multiple fleets

Agents with jobs, not just chat.

These four are a sample. We run multiple agent fleets — 50+ agents and growing every week — each owning a slice of operational busywork, working around the clock, and handing off to a human the moment a decision needs one.

  • reconciliation-agent on shift

    Three-way-matches POs, invoices, and receipts overnight — and surfaces only the breaks.

    Replaces
    the morning lost to match-break spreadsheets
    Returns
    ~32 hrs/wk · 96% matched touchless
    Guardrail
    proposes GL entries — never posts them
  • exception-triage-agent on shift

    Watches the exception queue 24/7, classifies each item, and resolves the routine ones.

    Replaces
    the 7am scramble to see what broke overnight
    Returns
    MTTR 4h → 18min · 89% auto-resolved
    Guardrail
    compliance & safety exceptions escalate with full context
  • catalog-qa-agent on shift

    Validates attributes, images, fitment, and allergens before anything publishes.

    Replaces
    line-by-line readiness checks before every launch
    Returns
    42% fewer publish defects · 18% faster
    Guardrail
    won't approve a publish gate — a merchant signs off
  • supplier-onboarding-agent on shift

    Normalizes incoming supplier files to your schema — units, enums, taxonomies.

    Replaces
    weeks of hand-mapping each new distributor feed
    Returns
    12M+ records normalized · 65% touchless
    Guardrail
    low-confidence mappings queue for human review

Four shown — 50+ agents across multiple fleets, and growing.

These cards are a sample, not a catalog. We run multiple agent fleets and train new ones for whatever your team does by hand — not just chat. Describe the busywork, and we’ll build a governed agent that does it around the clock. Want a custom agent? Yes — we build that.

Build a custom agent →
Governed AI, not prompt-and-pray

Agents that know when to step back.

Every response is intent-classified and confidence-scored before it reaches a user. High-stakes asks — medical, legal, financial — are blocked at the door and routed to a human. The full chain is written to an immutable audit trail.

◇ Confidence scoring ◇ Human-in-the-loop ◇ MCP tools ◇ Audit trail
Try the live agent Read the AI playbook →
Live · agent activity

Right now, agents are on the job.

This feed is our own fleet, working. When telemetry is reachable it streams real runs; when it's quiet or unreachable it falls back to a labelled sample — we never fake "live."

fleet · activity sample
  • reconciliation-agent matched 412 invoices · 3 flagged ✓
  • exception-triage routed SKU-mismatch → merchant queue ✓
  • catalog-qa blocked publish · 2 missing allergens ⛔→human
  • supplier-onboard normalized 1,840 records · 64% touchless ✓
  • reconciliation-agent 3-way match clean · 0 exceptions ✓
// streamed from the in-cluster agent fleet · falls back to a labelled sample
Model Context Protocol

We ship MCP, not slideware.

MCP is the open standard for exposing your tools, resources and prompts to AI agents — with governance built in and zero vendor lock-in.

architecture · 3 layers
  1. 01
    business · integration Your systems ERP, catalog, ops data — wired in once.
  2. 02
    ai · agent Any agent Claude, IDEs, custom orchestrators — your choice.
  3. 03
    mcp · governance MCP server One governed surface for tools, resources & prompts.
~/.config/mcp/client.json 
{
  "mcpServers": {
    "kettlelogic": {
      "url": "https://kettlelogic.com/mcp"
    }
  }
}

tools · 2
  • search_articles(query, limit=5) Find articles by slug or title.
  • get_industry_overview(industry) Plain-text overview for an industry.
resources · 3
  • kettlelogic://articles/manifest
  • kettlelogic://industries/list
  • kettlelogic://articles/{slug}

A live, hosted endpoint — point any MCP client at the config above, or self-host from mcp-kettlelogic on GitHub.

Is your business discoverable? → Read the MCP playbook See it on GitHub →
Live API surface

APIs you can actually call.

Real, documented endpoints — each with an OpenAPI spec, predictable shapes and a method you can read at a glance. Pick one on the left and send the request.

live base kettlelogic.com/api interactive docs kettlelogic.com/api/docs spec /api/openapi.json

GET /api/demo/ping
curl
response 200 · OK 
 

// example response · connect the API tier for live execution

 
 Swagger UI · kettlelogic.com/api/docs ↗ OpenAPI spec · /api/openapi.json ↗ 
 
 
 
 
 
     
  
 
  Live operations 

We run what we sell. 
 

We don't just claim SLO-driven ops — here's our own site. These figures are
live, read from the same in-cluster Prometheus that backs our
        Grafana; the full read-only dashboards for
kettlelogic.com are one click away,
        surfaced the same way we'd surface yours.

 
  
 
  status · kettlelogic.com live refreshed 30s 
 
     
  •  
     uptime · live  
     
      99.9%  
           kl-mcp targets, rolling 1h 
  •  
     api p95 latency  
     
      38ms  
           ingress p95, 5m 
  •  
     requests / min  
     
      —  
           ingress rate, 5m 
  •  
     error rate  
     
      0.0%  
           5xx ratio, 1h 
  •  
     health probes  
     
      3/3  
        
       
    •   startup 
    •   readiness 
    •   liveness 
      •  
       startup · readiness · liveness 
  •  
     MCP operations  
     
      —  
           mcp_operations_total 
  •  
     requests · 24h  
     
      —  
           ingress, kettlelogic hosts 
  •  
     build  
     
      passing
 
           main · ci passing 
    •  
  
 
 
    
 
 
 Outcomes, measured 

Real operations. Real numbers. 
 

Business command centers and data pipelines in production — and the
        movement they produced, measured the way operators measure.

 
 
 
 
 retail 
Shelf readiness in 6 weeks
 
 
 
 
 6wks 
 
to pilot
 
 
 42% 
 
fewer publish defects
 
 
 18% 
 
faster item launches
 
 
 
 
≈ 38hrs/wk  analyst time returned 
 
A regional retailer stood up a shelf-readiness control tower for item setup, images, and syndication — scorecards, exception queues, and launch gates tied to merchant and supplier workflows.
 
See the playbook   
 
 industrial manufacturing 
Supplier data normalization at scale
 
 
 
 
 12M+ 
 
records normalized
 
 
 65% 
 
touchless mapping
 
 
 2.3× 
 
team throughput
 
 
 
 
≈ 3FTE  of mapping work absorbed 
 
Multi-brand industrial catalog normalization pipelines with rule-based and AI-assisted mapping, readiness scoring, and remediation queues.
 
See the playbook   
 
 foodservice 
Menu and item governance for operators
 
 
 
 
 900+ 
 
locations
 
 
 30% 
 
faster menu changes
 
 
 Audit✓ 
 
ready approvals
 
 
 
 
≈ 120hrs/wk  store time saved 
 
Multi-location operator menu, modifier, and allergen control with tighter governance, approvals, and audit-ready change history.
 
See the playbook   
 
 
 
  
 
 
  
AI is an operations problem, not a shopping list. The teams that win don't buy the most GPUs — they fix their data, their workflows, and their governance first.
 
 Matthew Loschiavo Founder & CEO · author, “Before You Buy the GPUs” 
 
 
 
  
 
 
  Insights 
From the team that ships it.
 

Frameworks and field notes on governed AI, agents, and the economics of doing it well —
        the thinking behind what we build.

 All insights  
  
 
  
 
 
 
Building in the open
 

Proof, not promises 
 

The work is public. Every system below is a real repository you can
        clone, read, and run — shipped the same way we ship for clients.

 
 
  
  
  mploschiavo/mediaserver   
 
Self-hosted Jellyfin media server + full automation pipeline (Sonarr/Radarr/Lidarr/Readarr/Prowlarr/qBittorrent/Bazarr/Maintainerr/Jellyseerr) packaged as one declarative unit — deploy with one command on Docker Compose OR Kubernetes.
 
  A Python controller wires auth (Authelia SSO), an Envoy+TLS edge gateway, indexers, quality profiles, paths and subtitles — zero per-app clicks. The promises registry guarantees identical behavior across wipe-and-redeploy. 
 
 PythonAGPL-3.0Docker + k8s64 verified promisesReact 19 · Vite 6 · Tailwind v4 PWA 
 
 ★ —   Python  924 commits 
  
   
  
 
  
 
  
 
 
Let’s talk shop
 

Let’s build the system your
operations actually need. 
 

Custom software, intelligent workflows, and governed AI — designed
        around how your team really runs, not a template.

 
 Start a platform assessment Become a design partner 
 

Direct founder access · fixed-scope pilots · measurable outcomes