Controlled database access and security layer for AI agents
AI attention layer that learns what matters to you and surfaces priorities before they slip
Stage
Not given
MVP
Founded
Not given
Not given
Based in
ES
Not given
Pricing model
Freemium
Free
Pricing
Free tier: 30 successful calls per month. Pay Per Use: Managed AI €0.10 per question, BYOK €0.02 per question, Guard €0.01 per check. Enterprise: custom pricing.
Not given
Key features
Routes - define tables and columns agents can query with plain-language guides
Guard - check and approve every INSERT, UPDATE, or DELETE before execution
Read-only database connections with role-based access
REST API, MCP, and A2A protocol support
Support for Postgres, MySQL, MariaDB, and Supabase
Typed multi-question answers in one request
Data Map generation and review
Full query tracing and audit logs
BYOK - Bring your own LLM key from OpenAI, Google, Anthropic, or xAI
Connects to Gmail, Google Calendar, Slack and Google Meet
Shadow mode analyzes 30 days of past communications
Resolves people across channels into canonical contacts
Captures meeting decisions with reasoning and participants
Joins Google Meet as visible participant or captures from device
AI assistant Thara searches professional context and answers questions
Earns autonomy progressively through repeated approvals
Coverage honesty alerts when data sources fall behind
Categorizes attention into Needs You, Might Need You and Watching tiers
What makes it different
No data copying - agents query your database where it lives, data never moves
No code needed - set up routes, agents, and Guard rules through the dashboard without YAML or coding
Per-agent access control with clearcut boundaries
Exact rule enforcement with no model-based decisions at check time
Pay-per-success billing - you only pay for successful queries
Never claims certainty, only surfaces what may matter based on evidence
Learns around the way you work across different roles and professions
Maintains decision memory with reasoning and what changed
Requires explicit approval before acting on consequential items
Operates with read-only access and full user control over what is remembered