Archimetis

The systems read are named by class only, with no protocol, named product, connector catalog or write path.

The first party integration claim is one sentence, carried identically on /refining and /chemicals: "Connect DCS/historian data, integrity systems, incident logs, and process safety studies into a single model of your plant."

These classes reach below the application layer: a DCS is the plant control system and a historian is the time series system of record, which is harder to reach than a SaaS API.

Document classes are named elsewhere (Bowtie, LOPA and PHA studies; manuals; historical reports; incident notes), and the worked examples reference live data by instrument tag (EI-401, AI-101, TIC-7501.SP, K7502A), which only comes from a real historian connection.

The transport is described only in the negative: "Archimetis does not require software agents, inbound firewall rules, VPN tunnels, or changes to your OT network. Data is shared through customer-approved, outbound-only, cloud-native mechanisms." That tells a buyer what will not be asked of their network and nothing about how the connection is made. No protocol is named, whether OPC-UA, MQTT, a historian API, a file drop or a message bus.

It reads only. Nothing describes Archimetis writing into a DCS, a historian or a maintenance system, and the product is advisory by design.

Sourcearchimetis.ai/refining Integrate your plant and Enterprise-Grade Security card 03, /chemicalsread 2026-09-13

A real, documented, multistep sequence runs unattended, but it is the vendor's fixed method and customers cannot compose their own.

The vendor publishes five numbered steps. It reads the plant's existing Bowtie and LOPA documentation to find the highest consequence scenarios, checks each against live and historical operating data, and evaluates whether safety barriers are stressed during upsets, "using control performance, not alarms or trips." It then turns multisignal trends into an operating picture and surfaces the most likely risk signature and the next best checks.

The worked example runs it visibly: it searches documentation, extracts ranked scenarios, selects parameters, plots a month of tag data and interprets the margin, then "formulates and ranks hypotheses, validates them against process data." Generating hypotheses, validating them against data and ranking them is real multistep orchestration, not a single retrieval and answer turn.

Every step belongs to Archimetis. There is no workflow builder, step editor, exposed branching or conditional logic, named orchestration component, or anything a customer authors, saves, versions or reuses. The nearest customer control is "Engineers define the decision space. They set constraints, validate assumptions," which parameterizes an analysis the vendor designed.

Sourcearchimetis.ai/refining five-step How-to module and worked HDS example, /chemicalsread 2026-09-13

Reasoning is grounded in a maintained model of the customer's plant: "Connect DCS/historian data, integrity systems, incident logs, and process safety studies into a single model of your plant." A single model, built once and reasoned over repeatedly, is a maintained structure; the step is headed "Integrate your plant," and the artifact is the model, not the query.

The corpus is ingested deliberately: "Archimetis ingests plant documentation, manuals, historical reports, and incident notes," and the formal document classes are named: Bowtie, LOPA and PHA studies, the plant's approved risk framework.

The citation trail is explicit. "Reasoning is fully traceable. Every conclusion can be traced back to live data and plant documentation, making recommendations easy to inspect, challenge, and refine." The vendor states the grounding rule as a design constraint: "using the plant's approved risk framework, not a generic AI model."

Retrieval is shown as a discrete step. The worked example on /refining runs "I'll search our documentation for formal risk assessments, such as Bowtie or LOPA studies, that specifically cover the HDS unit," followed by a progress marker reading "Searching for HDS risk assessment documents," and the answers cite named equipment tags and documented consequence bands from those studies. That is a search against an ingested corpus, returning findings attributed to their sources.

No retrieval technology is disclosed, such as an index or embeddings.

Sourcearchimetis.ai/refining and /chemicals How we turn your data and expertise into decisions, Steer decisions with human judgmentread 2026-09-13

Oversight here is a surface the buyer operates, not only a posture. Under the heading "Steer decisions with human judgment": "Engineers define the decision space. They set constraints, validate assumptions, and inject real operating context into the analysis." Those are three named acts a person performs before the system reasons: bounding the space, approving the assumptions and supplying context.

The output path is review first by design, and the vendor states it as the product's defining choice: "AI + human expertise (not AI instead of humans)," and Archimetis "turns data into action, showing the reasoning behind every recommendation so engineers stay in control." The system produces "ready-to-act recommendations": ready to act, not acted. Nothing reaches the plant without a person.

The interaction asks for consent at each branch, and the worked examples show it. The agent does not proceed without a human choice: "Which path seems best to you," offering a plant wide search, a focused unit investigation or a specific hazard analysis; then "Which of these risks would you like to investigate further"; then "Which of these paths would you like to explore first" after three hypotheses are ranked. The fifth method step makes the restraint explicit: "Rather than issuing alerts, Archimetis frames the next best investigation... so engineers stay in control."

There is no approval queue, escalation path, risk tiering per action or sign off record, because there is no autonomous action to gate: the system never touches the controls.

Sourcearchimetis.ai/refining Steer decisions with human judgment, worked HDS example, How-to step 05, /chemicalsread 2026-09-13

There is an access surface and no attestation. Both sit under a section headed "Enterprise-Grade Security Built In," carried identically on /refining and /chemicals.

Access is covered. "SSO/OIDC identity integration" is named as a federation surface for customers, under a card headed "Zero-trust access that satisfies enterprise IT," alongside "least-privilege service accounts" and "no standing credentials."

An attestation is missing. No certificate, auditor or report is named anywhere, whether SOC 2, ISO 27001 or a penetration test summary, and there is no trust center to hold one. In its place is the phrase "meeting strict enterprise compliance expectations," which claims conformance to an unnamed standard verified by nobody. The customer facing surface is one item, single sign on through OIDC, with no role based access, SCIM or permission model over the agent, and the vendor does not name the gap.

The isolated GCP environment, TLS 1.3 in transit and KMS managed encryption at rest are infrastructure and tenancy properties, not access controls. The third card ("no software agents, no inbound firewall rules, no VPN tunnels, no changes to your OT network," with data moving through "customer-approved, outbound-only, cloud-native mechanisms") is a strong architectural posture for this buyer, but it is a design property, not a control the customer operates or a third party has verified.

Sourcearchimetis.ai/refining and /chemicals Enterprise-Grade Security Built Inread 2026-09-13

Two facts carry this. First, an explicit logging claim: "complete audit logs for every action," published in the security section. On a platform whose actions are agent actions, that is a record of what the agent did.

Second, the run can be inspected step by step in the product, and the worked example shows it.

The agent states its plan ("I will plot the following key parameters for the reactor over the past month," then five named tags), emits progress markers ("Searching for HDS risk assessment documents," "Assessing reactor temperature performance against safety limits"), renders the chart it reasoned over, and reports what it found, including the negative result: "While we haven't had an actual runaway event, operating this close to the safety limit reduces our ability to respond." A user can see which documents were searched, which tags were pulled, over what window, and how the conclusion followed.

This product's whole job is monitoring a refinery, and plant monitoring is not agent observability. What matters is the logging of the platform's own actions and the visible trace of the agent's reasoning.

What is missing is an observability surface documented as a product feature. There is no run history view, activity log a customer opens, trace export or admin console, and no statement of retention or who can read the audit logs. The audit log line sits in a card about smoothing internal security reviews, and the reasoning trace appears in a marketing walkthrough rather than as an artifact described for inspection.

Sourcearchimetis.ai/refining Enterprise-Grade Security card 02 and worked HDS example, /chemicalsread 2026-09-13

What Archimetis documents is a corpus and a learning claim, not agent memory. The site says it "turns ingested plant documentation and history into a living system that learns from every shift, decision, and outcome to preserve and scale institutional expertise." The ingested documentation is the corpus the system reads, not state it accumulated about its own work.

What remains is a claim that the model improves over time, with no named store, scope, read or write surface, or lifetime, and learning of that kind lives in the model rather than in a memory. "Preserve and scale institutional expertise" is the same claim in buyer language: knowledge that used to live in a retiring engineer's head now lives in the corpus.

No memory scope, lifetime or purge path is named. Nothing describes cross session recall, conversation history, saved context, a preference store, or the agent referring back to a prior run. The worked examples begin cold each time and set the scope again by asking the engineer "Which path seems best to you."

The plant model itself persists, but a persistent data structure built from customer documents is an application data model, not a memory layer.

Sourcearchimetis.ai/refining and /chemicals Learn from experienceread 2026-09-13

Each customer is isolated, but no location is named and there is no choice. The page states that each customer runs in a fully isolated GCP environment with VPC peered transfers, so operational data is never shared or mixed. Isolation is a tenancy property and residency is a location property, and the page answers only the first: a separate environment per customer says nothing about where it runs or whether the buyer may choose.

No region, residency commitment, self hosted option or selection surface appears anywhere. GCP is named as the vendor's own substrate, which is a subprocessor fact, not a deployment choice, and no residency is claimed at all.

Three items look like deployment and are not. VPC peered transfers are a private network path between clouds; they change how traffic reaches a tenant, not where the tenant is. "No software agents, no inbound firewall rules, no VPN tunnels, no changes to your OT network" describes what is not installed in the customer's plant, the opposite of a customer hosted option. TLS 1.3 and KMS managed encryption at rest are protection properties.

Sourcearchimetis.ai/refining Enterprise-Grade Security cards 01 and 03, /chemicalsread 2026-09-13

What the site shows is a coverage statement, not a catalog. Under "Archimetis adapts to a wide range of processes," the /chemicals page lists Bulk Chemicals, Ammonia, Petrochemicals, Specialty Chemicals, Polymers and Fine & Specialty, each with one line on what it reasons across: reforming, heat recovery, compression and synthesis; distillation columns, crackers and reactors; polymerization reactors and downstream processing; batch and continuous campaigns. "Adapts to" says one system covers many process types, not that a buyer selects among six things.

Ammonia is not a unit a buyer adopts, so dropping it would take nothing out of a purchase. It is a paragraph saying the same reasoning engine understands ammonia synthesis.

None of the marks of a pack is present: no page or URL per item, agent names, selection or enablement surface, marketplace, library, template gallery, install action or entitlement language. The industry pages (/refining, /chemicals, /industrials) are marketing segments of one product, referred to throughout in the singular: Archimetis reads, flags and recommends. The four benefit cards on /refining (Smarter Process Safety, Continuous Cost Optimization, Reliability Informed by Insight, Economic Performance at Full Potential) are outcomes the one system delivers, not units a customer adopts.

Sourcearchimetis.ai/chemicals Archimetis adapts to a wide range of processes, /refining Four ways we improve how your plant runsread 2026-09-13

Unattended operation is the product's premise, and the site shows it. "Unusual reactor conditions are detected over nightshift. Archimetis identifies possible causes and begins to triage and prioritize the actions required for operations. Engineers are in control, even before their morning meeting." Work done overnight with nobody logged in, presented on arrival, is the agent starting without a prompt.

Four distinct trigger classes are named. Schedule: "better decisions, from the first minute of every shift," with the shift boundary as the delivery cadence and start of shift as a named state. Data condition: "when the data looks alarming," with timestamps: "At 17:45, the WABT was detected to exceed its allowed operating window.

This continued into nightshift." Forecast threshold, which fires on a projection rather than a reading: "Fouling rate forecast: E4405 will exceed limit in 14 days," with the normalized fouling rate accelerating toward a 9.0 bar limit. Calendar state: "when turnarounds loom."

Findings carry clock times in the product UI (7:01 AM, 9:05 AM) and severity and ownership routing (High; Panel, Field Operations, Outside Operator, Maintenance), which is what a fired detection looks like, not a dashboard a person opens.

One line seems to read against this: "Rather than issuing alerts, Archimetis frames the next best investigation." That is about the form of the output, a framed investigation instead of a bare alarm, not about whether the system runs unprompted, and it sits in the same five step sequence that begins with continuous checking against live data.

No trigger configuration surface, webhook, external event subscription or schedule editor is documented; the trigger classes are the vendor's, not the buyer's.

Sourcearchimetis.ai/refining hero scenarios and How-to section, /chemicalsread 2026-09-13

The buyer does not choose the model. No model selector, admin entitlement over models, choice per request or per agent, disclosed routing policy, or bring your own model or key path appears on /refining, /chemicals or the homepage, and there are no tiers.

No provider or model family is named anywhere on Archimetis's own site. It says only "advanced AI," "AI that understands your plant" and "an AI partner." The one line touching the subject is about grounding, not selection: "using the plant's approved risk framework, not a generic AI model," which says the reasoning is constrained by customer documentation, not that a customer picks the engine.

Naming no provider is weaker disclosure than naming one, since a single named provider makes the absence of choice visible.

Sourcearchimetis.ai/refining, /chemicals, homepageread 2026-09-13

There is no developer surface. The site is small enough to list in full: the homepage is a video splash whose only outbound links are three industry pages (/refining, /chemicals, /industrials); the footer carries five destinations, all social or legal apart from a careers page; and /request-demo is reached from the single call to action. There is no docs or developer site, developer portal, API or GraphQL reference, SDK, CLI, MCP server, A2A card, webhook publisher, marketplace or app listing, and no partner or integrations program page.

There is no customer extension surface either. The nearest candidate is "Engineers define the decision space. They set constraints, validate assumptions, and inject real operating context," which parameterizes an analysis the vendor designed rather than adding capability. Nothing lets a customer add a connector, author an agent, define a trigger, or call the platform from outside.

Every documented interface points inward: data flows in through "outbound-only, cloud-native mechanisms" from the customer's side, and recommendations come back to people. No path exists for an external system to command Archimetis, query its state or build on it. The demo is the only gate that could hide a surface, and nothing behind it is documented.

Sourcearchimetis.ai homepage and site-wide footer, /refining, /chemicalsread 2026-09-13

Nothing puts a change to the agent under test: no harness before deployment, scored test cases, regression suite, golden set, A/B or shadow comparison of agent configurations, judge or rubric, accuracy, precision or hallucination metric, or published quality figure appears on any of the four pages.

It is easy to mistake the product's subject for its own testing. This product continuously evaluates a refinery: it assesses whether safety barriers are being stressed, validates hypotheses against process data, ranks causes by likelihood, and forecasts catalyst life. "Continuous cost optimization" and "the platform continuously analyzes energy use and process trade-offs" describe an optimization loop over the customer's plant, not over the agent.

Three claims are something else. That every suggestion "can be challenged and improved" is human review of each output. "Learns from every shift, decision, and outcome" is a learning claim with no mechanism. And "validates them against process data" is the agent checking its own hypothesis about the plant during a run, which is reasoning, not a test of the agent. The "custom ROI analysis for your operations" in the demo is a presales artifact people produce.

Sourcearchimetis.ai/refining Benefits and How-to sections, /chemicalsread 2026-09-13

No browser session the agent drives, desktop control, or remote or local computer control is documented, and the architecture rules it out. "Data is shared through customer-approved, outbound-only, cloud-native mechanisms," and the connection is to "DCS/historian data, integrity systems, incident logs, and process safety studies." Those are data paths into systems of record.

A DCS vendor could redesign its operator console entirely and Archimetis would be unaffected, because it never sees the console.

The vendor rules out the adjacent shape explicitly: "Archimetis does not require software agents, inbound firewall rules, VPN tunnels, or changes to your OT network." Nothing of the vendor's runs on the customer's endpoints.

The product's own chat and dashboard is Archimetis's client, not an interface it operates for someone. And the product replaces a person reading screens ("instead of scanning dozens of trends," "instead of asking humans to stare at more dashboards"), which removes the screen rather than driving it.

Process plants run decades old instrument software with no modern interface, and screen scraping is a common answer there. Archimetis takes the other route: it reads the historian directly and never touches the control network.

Sourcearchimetis.ai/refining Integrate your plant and Enterprise-Grade Security card 03, /chemicalsread 2026-09-13