generated: '2026-07-21'
method: searched
source: https://docs.sphinx.ai
docs: https://docs.sphinx.ai
description: >-
  Cross-cutting conventions for working with Sphinx as a developer. Sphinx is a
  data-science AI agent (VS Code / Cursor extension + CLI + in-notebook Python
  library) rather than a REST API, so these conventions describe the developer
  contract of the agent platform, not HTTP request/response semantics. No public
  OpenAPI is published, so pagination / idempotency HTTP conventions do not apply
  and are recorded as not-applicable rather than fabricated.
authentication:
  style: api-key + jwt
  api_key_env: SPHINX_API_KEY
  api_key_prefix: "sk-"
  interactive_login: "sphinx-cli auth login (30-day token in ~/.sphinx)"
  ref: authentication/sphinx-authentication.yml
secrets:
  description: >-
    Secrets and credentials are managed centrally and accessed in notebooks via
    the Sphinx Library (get_user_secret_value, get_connection_credentials); code
    execution happens on the user's own kernel.
  docs: https://docs.sphinx.ai/configuration/secrets
connections:
  description: >-
    Cloud data warehouses/services (Snowflake, Databricks, Google Colab, Azure
    ML, AWS SageMaker, JupyterHub) connect through Integrations; credentials are
    retrieved securely at runtime.
  docs: https://docs.sphinx.ai/configuration/integrations
mcp:
  role: client
  description: >-
    Sphinx acts as an MCP client — it connects external tools and services to the
    agent via the Model Context Protocol. It does not publish a hosted MCP server.
  docs: https://docs.sphinx.ai/configuration/mcp
agent_modalities:
  description: >-
    Four interaction modes governing autonomy: Agent (autonomous), Plan (approve
    strategy first), Safe Mode (approve every code execution), and Streamlit
    (build interactive apps).
  docs: https://docs.sphinx.ai/agent/agent-modalities
guardrails:
  description: Configurable permission levels controlling what code Sphinx generates and executes.
  docs: https://docs.sphinx.ai/configuration/guardrails
memory:
  description: >-
    A semantic cache ("Memories") captures user preferences, data relationships,
    and organizational knowledge; toggleable per run via --no-memory-read /
    --no-memory-write.
  docs: https://docs.sphinx.ai/configuration/memories
idempotency:
  supported: false
  note: No idempotency-key contract is documented (not a REST transaction API).
pagination:
  applicable: false
error_envelope:
  applicable: false
  note: No public HTTP API error catalog is published.