Skip to content

Cloud vs Local OSS

SpectraSherpa is one scientific workbench with two practical entry points. The workflow concepts are shared; account policy, demo limits, and centrally managed AI behavior belong to hosted deployments.

Question SpectraSherpa Cloud Local OSS
Best first use Fast browser evaluation, managed users, hosted demo workflows Local data evaluation, source inspection, plugin development
Install None for the user pip install spectra-sherpa
Data limits Demo and enterprise policy may limit uploads, runs, or AI use No hosted demo limits; constrained by your machine and local policy
AI setup Centrally configured by the deployment; demo users do not bring keys Optional BYO OpenAI-compatible, Anthropic, or Ollama chat provider
Extension User-facing cloud docs only Full source, node plugins, exports, providers, and tests

SpectraSherpa Cloud

Use Cloud when you want a browser-first managed environment with accounts, demo access, centrally configured AI features, and no local Python setup. It is the right first path for an enterprise evaluator who wants to try FTIR, NIR, Raman, or UV-VIS workflows quickly.

The public demo is free to try at demo.spectrascientific.ai. Create an account with access code welcome_to_spectra_sherpa. Public demo policy is documented on the hosted site at Demo Access and Limits.

Cloud documentation is written for users. It explains how to run workflows, understand demo limits, use Advisor and Guidance, and review outputs. It does not teach service development or deployment internals.

Hosted feature policy

Cloud profiles can differ. The public demo, enterprise trial, and paid enterprise deployment may have different upload limits, AI configuration, retention policy, HITRAN key handling, and sharing behavior.

Local OSS

Use local OSS when you want to run SpectraSherpa on your own computer, inspect or modify the source, build plugins, or work without a hosted account.

Local OSS provides:

  • local FastAPI/Vue application
  • local SQLite-backed app data
  • workflow builder and node execution
  • core numpy/scipy/scikit-learn chemometrics
  • bring-your-own-provider chat configuration for OpenAI-compatible, native Anthropic Messages, or Ollama endpoints
  • optional spectra-sherpa[scp] support for SpectroChemPy-backed readers and nodes
  • optional spectra-sherpa[hitran] support for HITRAN/HAPI synthesis

Start with 30 Minutes to Local Compute, then check Supported File Types before importing instrument files.

AI Use and BYOK

SpectraSherpa separates scientific computation from AI assistance. PCA, PLS, SIMCA, KNN, MCR, preprocessing, and most plots run as local scientific application code. Reproducibility depends on the recorded workflow, data, environment, seed, and documented numerical tolerances. LLM features are optional assistance layers around interpretation, writing, workflow guidance, and selected suggestion tasks.

Local OSS can use a bring-your-own-provider chat endpoint configured by environment variables or the app's local BYO Chat settings when enabled. The built-in local transport supports OpenAI-compatible /chat/completions APIs (including OpenAI and compatible hosted providers), Anthropic's native Messages API, and Ollama's local OpenAI-compatible endpoint. Select the provider explicitly with CHAT_ENDPOINT_PROVIDER; OpenAI-compatible and Anthropic providers require a key, while Ollama does not.

Private and loopback model endpoints are blocked by default. An operator who intentionally runs a trusted local model server can enable them for the configured provider with CHAT_ENDPOINT_ALLOW_PRIVATE=true; review the endpoint carefully because this relaxes the SSRF protection for that chat destination. SPECTRA_SHERPA_ALLOW_PRIVATE_LLM_ENDPOINTS=1 remains a process-wide compatibility override and is broader than the per-provider setting.

Cloud can be configured by Spectra Scientific for managed LLM access, BYOK access, or both, depending on the enterprise/demo profile. BYOK means your organization is also depending on the LLM supplier for model availability, usage limits, pricing, data-handling terms, and service reliability.

Scientific Caution for AI Output

Treat AI output as a labeled draft or suggestion, not a measurement. LLMs can be plausible and wrong, can miss sample-ID or validation-design problems, can overstate weak evidence, and can change behavior when an upstream provider updates a model.

AI-generated or AI-assisted artifacts in SpectraSherpa are labeled in the product. The main impacted features are:

  • Sherpa Advisor chat responses
  • Ambient Guidance suggestions
  • AI report narratives and data stories
  • peak-identification or vibrational-assignment suggestions when enabled
  • workflow or code-generation suggestions when enabled

Review AI output against spectra, metadata, file provenance, validation splits, plots, metrics, lab records, and domain knowledge before using it in scientific or customer-facing decisions.