Skip to content

Latest commit

 

History

History
120 lines (97 loc) · 7.82 KB

File metadata and controls

120 lines (97 loc) · 7.82 KB

Capabilities

This guide helps you evaluate DataFoundry's capability scope. After reading it, you can tell what Web, TUI, and API each support, and which features depend on backend capabilities or external resource configuration.

Status is based on current code:

Status How to verify
Ready to try After local startup, with a model key configured, the built-in DTC Growth Review runs end to end.
Requires configuration Feature entry exists but needs a model key, database credentials, files, MCP server, or Skill package.
Capability-controlled Read GET /api/v1/capabilities and enable or hide related entry points from the response.
Password auth boundary Cookie-based password sessions cover account registration, login, reset, and CSRF. Production deployments still need secret management, audit export, access control policy, and operations monitoring.

Overview

Capability Web workbench TUI Backend/API How to verify
Natural-language data analysis Ready to try Ready to try Ready to try Configure an LLM key and ask questions with dtc-growth-demo.
Built-in DTC growth case Ready to try Ready to try Ready to try Data source list includes DTC Growth Review; each user gets a read-only workspace copy.
Data source registration and test Ready to try Select configured sources Ready to try GET /api/v1/datasource-types, POST /api/v1/datasources/:id/test.
Schema fetch and table preview Ready to try Via agent tool results Ready to try POST /api/v1/datasources/:id/introspect, GET /schema, GET /tables/:table/preview.
Read-only SQL analysis Ready to try Ready to try Ready to try Agent run inspects schema first, then runs queries through tools.
Model configuration Requires configuration Uses server model config Requires configuration .env or /api/v1/model-profiles.
Model connection test Ready to try Uses server model config Ready to try POST /api/v1/model-profiles/:id/test.
Analysis trace Ready to try Ready to try Ready to try View steps, tool calls, run events, and SQL audit.
Semantic Trace DAG Ready to try No graph view Ready to try Open the Web trace graph or GET /api/v1/sessions/:id/trace-dag.
Artifact outputs Ready to try View session outputs Capability-controlled artifact.list, artifact.export, artifact.promote.
Session history Ready to try Resume with /resume Capability-controlled conversation.memory, conversation.title.
Concurrent sessions and queued prompts Ready to try One active terminal flow Ready to try Start runs in separate sessions; submit another prompt during an active Web run.
Checkpoint branches Ready to try No branch controls Ready to try Re-ask from an earlier turn or POST /api/v1/sessions/:id/branches.
Evidence references Ready to try No selection UI Ready to try Reference an output or selection and inspect evidenceRefs diagnostics.
Data Link graph Requires configuration No graph view Requires configuration Configure a compatible Data Link MCP server, then open Data Link.
User identity Password login / register screens Uses backend session identity Ready to try GET /api/v1/me, /api/v1/auth/*.
Workspace files Upload, view, download, delete, reuse Use enabled files via run_config Capability-controlled files, GET/POST /api/v1/files, POST /api/v1/files/:id/promote.
Chat attachments Ready to try No attachment upload command Capability-controlled chat.fileUpload, POST /api/v1/chat/uploads.
Image input Input controlled by switch No image input command Capability-controlled chat.imageInput.
Knowledge bases Requires configuration Enabled resources via run_config Capability-controlled knowledge, kb.chunking, kb.citationPolicy.
MCP tools Requires configuration Enabled resources via run_config Capability-controlled mcp, mcp.stdio, mcp.toolPolicy.
Skills Requires configuration Select with /skill Capability-controlled skills, skill.resourceBinding.
Cancel run Ready to try No slash command Ready to try POST /api/v1/runs/:id/cancel.

Backend capability keys

GET /api/v1/capabilities returns the keys below. Clients use them to control UI, run configuration, and resource entry points:

Key Controls
artifact.export Artifact export.
artifact.list Session artifact list.
artifact.promote Promote file artifacts into the workspace.
chat.fileUpload Chat attachment upload.
chat.imageInput Image input.
conversation.memory Server-side session memory.
conversation.title Session title persistence.
interaction.resume Human interaction resume after refresh or session switch.
datasource.fieldMasking Data source field masking configuration.
datasource.extendedTypes Extended data source types.
datasource.introspectionPolicy Schema introspection policy.
datasource.queryPolicy Query row limit, timeout, and write-deny policy.
datasource.samplePolicy Sample preview policy.
datasource.server Server database connection fields.
files Workspace file assets.
kb.chunking Knowledge base chunking configuration.
kb.citationPolicy Knowledge base citation policy.
kb.scope Knowledge base scope.
llm.advancedSampling Advanced model sampling parameters.
llm.samplingParams Model sampling parameters.
knowledge Knowledge resources in runtime.
mcp MCP resources in runtime.
mcp.stdio stdio MCP server configuration.
mcp.toolPolicy MCP tool policy.
skill.resourceBinding Skill resource binding.
skills Skill resources in runtime.

Web workbench

The Web workbench suits local demos and daily analysis:

  • Left panel: sessions and workspace resources.
  • Center: conversation, step cards, and human confirmations.
  • Right: overview, semantic trace, outputs, evidence-aware step details, and workspace files.
  • Input box: model selection, resource toggles, @ mentions, attachments, evidence chips, queued prompts, and stop run.
  • Session list restores history via server /api/v1/sessions; earlier turns and checkpoints can create persistent branches.
  • Data Link opens a workspace graph when a compatible MCP service is configured.

See Web workbench guide.

TUI

The TUI suits remote servers and terminal workflows:

  • Chat-first terminal UI with a separate /outputs page.
  • /datasource to select a data source.
  • /skill to select a Skill.
  • /resume to restore server session history.
  • Password sign-in with a local session cache (requires a running API).
  • Tab completion, input history, and Chat view scrolling.

Registered commands are defined in TUI guide.

API and integration

The backend exposes two entry types:

Entry Purpose
POST /api/copilotkit Start an agent run and return an AG-UI event stream.
/api/v1/* Manage resources, files, sessions, outputs, and configuration.

Integrators should manage resources through the configuration API and start analysis through Agent Runtime. Data source credentials are submitted only when creating or updating resources.

Security boundaries

  • Clients must not put database passwords, model API keys, or MCP tokens in the agent run body.
  • Read APIs do not return plaintext credentials.
  • SQL execution applies read-only limits, row limits, timeouts, and audit.
  • All product paths require an authenticated password session; there is no anonymous or development-token identity.
  • Production deployment still needs secret management, audit export, access control policy, and operations monitoring.

Continue with Security.