SVAJNA

Autonomous data science with verifiable execution.

SVAJNA turns local data into reproducible evidence: it profiles datasets, detects quality risks, records an event trail, and saves reports and analytical memory for later comparison.

npm install
npm run build
node packages/cli/dist/index.js init
node packages/cli/dist/index.js analyze ./sales.csv
node packages/cli/dist/index.js report
node packages/cli/dist/index.js status

Analysis never changes the source dataset. It writes reproducible state to .svajna/.

Build first, then configure an MCP client to launch the local, project-bounded server:

node /absolute/path/to/SVAJNA/packages/mcp/dist/index.js

The server provides data_profile, data_compare, analysis_execute, and memory_read, plus a svajna://capabilities resource. Dataset paths must be relative to the MCP server's working directory; paths outside that project are rejected.

The advanced runtime primitives and continuation plan are recorded in docs/architecture.md and AGENT_HANDOFF.md.

See AGENT_HANDOFF.md for the current phase, dependency list, and continuation notes.