AssistantSpec

Overview

AssistantSpec is the org-wide configuration file that controls how lakecode behaves across your entire team. It defines which workflows are available, what defaults apply, and what guardrails are enforced.

AssistantSpec is stored in Unity Catalog and applied automatically to every lakecode session in your organization.

Configuration file

AssistantSpec is defined as a YAML file and pushed to your workspace using the deployment CLI:

# assistantspec.yaml
version: 1
org: acme-corp

# Default workspace settings
defaults:
  catalog: prod
  schema: analytics
  warehouse: lakecode_wh
  timezone: America/New_York

# Available workflows
workflows:
  - prove        # Evidence packs
  - debug        # Job debugging
  - profile      # Table profiling
  - query        # Natural language SQL
  - migrate      # Schema migrations

# Disabled workflows (hidden from users)
disabled_workflows:
  - drop         # Disable destructive operations

# Naming conventions
naming:
  table_prefix: ""
  staging_schema: staging
  archive_schema: archive
  evidence_schema: lakecode_evidence

Applying configuration

# Validate the spec
lakecode-deploy spec validate assistantspec.yaml

# Push to your workspace
lakecode-deploy spec apply assistantspec.yaml

# View the active spec
lakecode-deploy spec show

Configuration changes take effect immediately for all new sessions. Active sessions continue with their original configuration until they complete.

Approval policies

Approval gates are configured within AssistantSpec. See Approval Gates for the full reference.

approval:
  read: auto
  write: require
  destructive: require
  destructive_reviewers: 2
  exempt_catalogs: [sandbox, dev_scratch]
  strict_catalogs: [prod, analytics]

Model configuration

Control which LLMs lakecode uses and how requests are routed through AI Gateway:

models:
  # Primary model for workflow planning
  planner: claude-sonnet-4-20250514

  # Model for SQL generation
  sql: claude-sonnet-4-20250514

  # Model for evidence summarization
  summarizer: claude-haiku-4-5-20251001

  # AI Gateway endpoint (Databricks)
  gateway_endpoint: /serving-endpoints/ai-gateway

  # Rate limits per user per hour
  rate_limit: 100

  # Max tokens per request
  max_tokens: 8192

User permissions

Control what individual users or groups can do:

permissions:
  # Admins can modify AssistantSpec and approve all operations
  admins:
    - admin-group@acme.com

  # Reviewers can approve operations but not modify config
  reviewers:
    - data-eng@acme.com
    - platform-team@acme.com

  # Users can trigger workflows but need approval for WRITE/DESTRUCTIVE
  users:
    - analysts@acme.com
    - data-science@acme.com

  # Restrict specific catalogs to specific groups
  catalog_access:
    prod: [data-eng@acme.com, platform-team@acme.com]
    analytics: [analysts@acme.com, data-science@acme.com]
    sandbox: ["*"]  # Everyone

Connector configuration

Configure Slack and Jira integrations within AssistantSpec. See Connectors for setup details.

connectors:
  slack:
    webhook_url_secret: lakecode/slack-webhook
    default_channel: "#data-ops"
    approval_channel: "#data-approvals"

  jira:
    base_url: https://acme.atlassian.net
    api_token_secret: lakecode/jira-token
    default_project: DATA
    evidence_label: lakecode-evidence

Full reference

For the complete list of AssistantSpec fields, defaults, and validation rules, run:

lakecode-deploy spec reference