Skip to content

AI integrations

FlaskTrack can use organization-configured AI providers for assisted authoring, knowledge retrieval, reporting, molecular-design features, and the Digital Lab Assistant.

Organization AI provider settings
Administrators configure approved AI providers and organization-level integration settings.

Provider choices depend on the deployment and can include hosted or self-hosted model endpoints.

Configure a provider

An authorized administrator supplies the provider credential, endpoint, and model configuration required by the selected integration.

Treat provider credentials as production secrets and restrict who can change them.

For a self-hosted Ollama-compatible endpoint, the FlaskTrack server must be able to reach the configured API address. localhost refers to the FlaskTrack container or host itself, not an administrator's laptop.

Digital Lab Assistant

The Digital Lab Assistant can:

  • answer questions using available FlaskTrack and organizational context;
  • discover permitted laboratory records;
  • prepare controlled multi-step action plans;
  • ask for clarification when a real user decision remains;
  • wait for user approval before executing planned mutations.

See Digital Lab Assistant.

Supported AI use cases

Depending on enabled modules, AI can assist with:

  • drafting protocols and workflows from prompts or documents;
  • identifying candidate catalog items required by an imported procedure;
  • answering questions from organization-approved notes and files;
  • preparing reviewed FlaskTrack action plans;
  • drafting SQL for custom reporting;
  • drafting Data Pipeline code;
  • creating molecular sequence scaffolds or design suggestions.

Built-in assistant versus external agents

The organization AI provider powers FlaskTrack-managed AI features.

External agents are separate integrations that authenticate to FlaskTrack through organization-scoped API credentials and discover the registered tool interface.

See External AI agents and MCP tool calling.

Human review is mandatory

AI output can be incomplete, scientifically incorrect, internally inconsistent, or unsuitable for a specific laboratory.

Generated scientific entities should enter the applicable draft or reviewable process.

An authorized person must verify the exact:

  • procedure;
  • units;
  • timing;
  • materials;
  • hazards;
  • forms;
  • inventory effects;
  • record selections;
  • compliance controls.

For action plans, review every proposed mutation before approval.

Action-plan controls

A proposed AI action plan does not execute simply because the model produced it.

FlaskTrack applies:

  • typed record validation;
  • organization scoping;
  • user permissions;
  • tool input validation;
  • action review;
  • runtime execution checks;
  • compliance and electronic-signature rules.

See AI action plans.

Data handling

Only submit data that the organization is authorized to process with the selected provider.

Provider-hosted processing is governed by the organization's provider account and terms.

Self-hosted models can reduce external transfer but still require:

  • access control;
  • logging;
  • patching;
  • network security;
  • capacity management;
  • backup and incident procedures where applicable.

Knowledge sources

Files and notes explicitly selected for an organizational knowledge base may be used to provide context to supported AI features.

Keep sources current, remove superseded documents, and do not treat retrieved text as automatically approved instruction.

Troubleshooting

Authentication error

Verify the provider credential and provider-account permissions.

Model not found

Confirm that the configured model is available to the provider account or self-hosted endpoint.

Connection refused

Verify network access from the FlaskTrack deployment to the configured provider.

Invalid structured output

Retry, reduce prompt complexity, or complete the task manually. Never approve malformed or scientifically suspect output merely to continue.

Unexpected record choice

Discard the plan and refine the request or clarification. Do not approve a plan using the wrong FlaskTrack entity.

Unexpected content exposure

Disable the integration if necessary, rotate affected credentials, preserve audit evidence, and follow the organization's incident procedure.