Comparison guide · as of 2026-10

Matflow vs. Alchemite — an honest comparison

Alchemite is a sparse-data machine-learning engine for materials and formulation R&D. We cannot verify its model behaviour, accuracy or pricing from this repository, so this page does not assert them. It shows the verifiable Matflow path for sparse data — synthesis, evaluation, prediction with intervals — and the questions to ask both systems. As of 2026-10.

Comparison policy

What this page does — and does not — claim about Alchemite (Intellegens)

Two commitments, both about trust.

We do not assert Alchemite (Intellegens)'s capabilities.Alchemite (Intellegens) is a sparse-data machine-learning engine for materials and formulation R&D. Product capabilities, licence terms and pricing change without notice, and we cannot independently run another company's software from this repository. Every competitor fact on this page is either marked as unverified or phrased as a question to ask. As of 2026-10 — re-verify before relying on it.

We do show our own work.Every Matflow dimension below points to the mechanism in this repository — the module that labels evidence, the endpoint that serves prices, the code that validates storage paths. Audit the claim instead of taking a badge's word for it.

What Matflow verifiably ships

Five dimensions you can check in the repository

These are the axes we would use to evaluate any research platform — and the ones we hold ourselves to.

Evidence

Evidence classes and engine cards

Every scientific response carries one of six evidence classes — MEASURED, COMPUTED, PREDICTED, EXTRACTED, HYPOTHESIS, DEMO — with a level, plus an engine card naming what produced it. Surrogate paths are relabelled DEMO rather than printed as fact (backend/core/engines.py:114, 251-329).

Pricing

Transparent self-serve pricing

A free tier and published plans anyone can reach without a sales call. The pricing page renders the same catalog the server enforces through GET /api/billing/plans (client/src/pages/StudioPricing.jsx, backend/core/pricing.py:1-48).

Optimization

Cost-aware optimization

Cost is a first-class objective: TEA prices raw materials, equipment (six-tenths rule), CapEx/OpEx and end-of-life credit, and screening-grade LCA adds GWP, energy demand and E-factor — so candidates are ranked by net value, not performance alone (client/src/pages/docs/content.jsx, stages/catcost and lca/overview).

Ownership

Own cloud, no lock-in

Data lives under a per-account directory with every served path validated (backend/core/paths.py:71-105); datasets export as CSV and RO-Crate v1.1 provenance ZIPs, and the repository ships a single-tenant Docker Compose overlay (deploy/production/docker-compose.single-tenant.yml).

Access

Full-module access

All plans include the full pipeline — the differences are throughput, not features (client/src/pages/docs/content.jsx, account/billing). No module is held back behind an enterprise quote.

Side-by-side, honestly bounded

Matflow vs. Alchemite (Intellegens): verifiable dimensions and open questions

The middle column is enforced in this repository. The right column states what we could not verify and what to ask — never what the other platform does or cannot do.

Dimension	Matflow (verifiable)	Alchemite (Intellegens): what we could not verify — ask
Trying it before sales	Free tier and self-serve signup; the full pipeline runs on a bundled demo dataset before you upload anything (client/src/pages/docs/content.jsx, getting_started/overview).	Demo-first sales is common in this segment. We could not verify a self-serve trial from public sources as of 2026-10. Ask: can I run a trial on my own file before procurement?
Published pricing	Monthly and annual plans with metered limits and credit packs, rendered live from the billing catalog the server enforces (backend/core/pricing.py:1-48).	Quote-gated for most enterprise platforms; not public. Ask for an itemized, metered price sheet and what resets when.
Evidence labelling	Six classes with levels and an engine card on every result; ML and surrogate outputs are PREDICTED/DEMO, never presented as measurements (backend/core/engines.py:114, 251-329).	Varies; not publicly documented at result level. Ask whether measured, computed and predicted outputs are distinguishable on the result itself.
Cost and sustainability as objectives	TEA (raw materials, six-tenths equipment sizing, CapEx/OpEx, end-of-life credit) and screening-grade LCA (GWP, energy demand, E-factor) can join the optimization objectives (client/src/pages/docs/content.jsx, stages/catcost; lca/overview).	Varies; costing often sits in a separate module or a consulting engagement. Ask whether cost can be an optimization objective alongside performance.
Data ownership and export	Account-scoped storage with path validation; datasets export as CSV and RO-Crate v1.1 provenance ZIPs, with training files SHA-256 hashed (backend/core/paths.py:71-105; backend/core/provenance.py).	Varies. Ask for a portable export specification — format, provenance and deletion path — before signing anything.
Deployment model	Browser-first web app; a single-tenant Docker Compose overlay ships in the repository, and the readiness endpoint reports tenancy, region gating and BYOK flags without exposing secrets (deploy/production/docker-compose.single-tenant.yml; backend/core/soc2.py:367-411).	Varies; on-prem and private-cloud options are usually an enterprise negotiation. Ask where the software runs and what single-tenant means in the contract.
Module access	Every plan includes the full pipeline: synthesis, evaluation, prediction, optimization, TEA, LCA, screening, DOE and the physics engines — limits are throughput, not features (client/src/pages/docs/content.jsx, account/billing).	Varies; capabilities are often tiered or licensed per module. Ask for the full module list and which tier unlocks each one.
Programmatic surface	OpenAPI 3.0 spec and interactive reference at /developers, a Python SDK under tools/sdk, hashed API keys, and HMAC-signed webhooks with retries (client/src/pages/docs/content.jsx, sdk/overview; backend/core/webhooks.py:1-14).	Varies; API access may be limited or enterprise-only. Ask for an OpenAPI spec, SDKs, key lifecycle and webhook events.
Reproducibility	Benchmark reports ship downloadable reproducibility bundles, and RO-Crate exports carry provenance and engine cards (client/src/pages/StudioBenchmarks.jsx; backend/routes/rocrate_routes.py).	Not public. Ask whether a reviewer can reproduce a result from the dataset version, configuration, seeds and metrics.

As of 2026-10Matflow claims are shipped features with in-repo backing; competitor capabilities are intentionally absent. Re-verify all third-party facts before relying on them.

Market context

Published pricing is the exception, not the rule

Price transparency in this category is rare — which is exactly why we publish ours.

What was publicly verifiable as of 2026-10.Most platforms in this segment are quote-gated. The published price points we could verify at the time of writing were ChemCopilot at $219 per user per month and Mat3ra at $360 per year plus compute. Those figures are examples of market context, not comparisons to Alchemite (Intellegens), and they may have changed — re-verify before citing them. We could not verify published pricing for Alchemite (Intellegens) as of 2026-10.

See Matflow's published plans

FAQ

Questions we expect from Alchemite (Intellegens) evaluators

Direct answers, including the limitations.

Is this a model-accuracy comparison against Alchemite?

No. Matflow publishes no head-to-head benchmark against Alchemite because we cannot independently run another vendor's engine from this repository. Our benchmark results are on public datasets with downloadable reproducibility bundles (/benchmarks, /benchmarks/methodology). Accuracy on your data is a trial question, not a page claim.

How does Matflow approach sparse data?

Data Synthesis expands a small dataset, then Evaluation stress-tests the generated rows with 20+ metrics before they are trusted in modeling (client/src/pages/docs/content.jsx, stages/synthesis; stages/evaluation). Prediction adds conformal intervals and an extrapolation flag (stages/prediction). We do not claim this approach matches Alchemite's.

What evidence label do Matflow predictions carry?

PREDICTED with a model card, and surrogate or fallback paths are relabelled DEMO rather than printed as facts (backend/core/engines.py:114, 251-329). The extrapolation flag marks predictions outside the training envelope.

How does pricing compare?

Matflow publishes plans and metered limits at /pricing. Most platforms in this segment quote privately, and we could not verify Alchemite's pricing from public sources as of 2026-10 — so no price comparison is made here.

Can I take the trained model with me?

Models are per-owner rows in the registry; the Model Hub exports a serialized model, and datasets plus provenance export as RO-Crate (client/src/pages/docs/content.jsx, models/hub; export/overview). There is no proprietary file format holding your results hostage.

Does this page claim Matflow is better than the other platform?

No. We cannot independently run or verify another company's product from this repository, so there is no feature-by-feature matrix and no superiority score. This page shows the dimensions of Matflow we can point to in code, and the questions to put to both platforms — ours included — during your own evaluation.

How were the Matflow claims on this page verified?

Each claim in the comparison table names the module, endpoint or documentation page that backs it. The public benchmark numbers live at /benchmarks with methodology at /benchmarks/methodology, and the broader evaluation framework is at /alternatives.

Judge it on your own data

No demo gate and no sales qualification: sign up, load the bundled dataset or your own file, and inspect every result's evidence class before you decide. The comparison that matters is the one you run yourself.