The intelligence layer
for AI agents
Connect data sources. Manage context windows. Observe everything. Without the chaos.
All the context that flows from your data sources builds a universal memory you can search at any time.
Immediate awareness. Auto-compacted summaries of live context. Always included in every query.
Working knowledge via Activation Beacon. ~85% compression with full semantic fidelity.
Deep archives. Auto-rollover to vector storage for semantic search across history.
How Datagran Solves Context
AI agents need more than prompts. They need the right knowledge at the right time, managed context that doesn't overflow, and visibility into every decision they make.
Plug Knowledge
Connect your AI to the data sources it needs—ads platforms, CRMs, databases, files—in one click. No custom integrations. No token management headaches.
Manage Context
Three memory layers that work like a human brain—short-term for live context, mid-term for compressed knowledge, long-term for infinite history.
Observe Everything
See every decision your AI makes in real-time. Track what context was used, what actions were taken, and why. Full audit trails for compliance and debugging.
See Beacon Compression in Action
We scraped the Wikipedia Paris article using Firecrawl. Watch how 267,065 tokens compress into 46,080 beacon KV positions—~6x KV cache reduction with sub-second query latency.
## Paris
Paris is the capital and largest city of France, with an official estimated population of 2,102,650...
### History
- Founded: 3rd century BC (as Lutetia)
- Population: ~2.1 million (city), ~12 million (metro)
- Landmarks: Eiffel Tower, Louvre, Notre-Dame...
- Economy: Tourism, finance, fashion, art...
How Beacon Compression Works
Your document's 267K tokens are fed into the Activation Beacon model chunk by chunk (8K tokens per chunk).
With compression ratio α=8, tokens compress into beacon KV activations—resulting in ~6x memory reduction.
The model uses the 46K beacon KV cache instead of the full 267K—sub-second latency on warm GPU.
AIC (AI Interaction Controller)
The enforcement engine that ensures every AI action is safe. Risk scoring, policy checks, approval flows, and complete observability—all in one place.
Create a new campaign with $5,000 budget
This request exceeds the auto-approval limit ($1,000). I've flagged this for human review before execution.
1. Risk Scoring
Evaluates context, intent, and historical patterns.
2. Policy Enforcement
Checks against budget limits, allowed actions, and targets.
3. Observability
Logs full trace, including prompt, response, and API payload.
Get Started in Minutes
Build AI agents with persistent memory. Query your brain first, fetch only what's missing.
Create Account
Sign up at the Partner Portal and grab your API keys in seconds.
Connect to any of our Integrations
Connect to OAuth data sources and more with a click.
Connect Your AI
Use our MCP server for Claude/Cursor, or call our REST API directly.
Enable Memory
Toggle auto-compile in the Partner Portal to retain data as persistent brain memory.
Query the BrainKEY
Call /api/context/brain first to check if data already exists in memory.
Fetch if Empty
No memory? Fetch from integrations. Data auto-compiles to the brain for next time.
Result: Your AI only fetches data once. Every subsequent query hits the brain—instant answers, lower costs, persistent memory across sessions.
Case Studies
See how enterprises are securing their AI
Marketing AI with Full Observability and Control
Challenge
Needed a way to safely give AI agents access to marketing systems without risking campaign tampering, data leaks, or untraceable decisions.
Solution
Implemented Datagran as middleware between AI agents and ad platforms (Meta, Google Ads). Provided a secure proxy, real-time observability, and automated performance reporting.
Results
- 100% visibility into all AI-initiated ad actions
- Improved ad performance through automated content learning
- Full audit trails for marketing ops and leadership
- Reduced risk of unauthorized changes
Securing AI Agents in Government Customer Support
Challenge
Deploying AI agents in sensitive support roles requiring access to CRMs and APIs, while meeting strict government compliance and audit standards.
Solution
Integrated Datagran as a control layer to enforce fine-grained action control, real-time monitoring, and risk evaluation for every API call.
Results
- Met internal/client security requirements for AI
- prevented unauthorized or destructive calls
- Enabled broader AI rollout in sensitive workflows
- Built trust through transparent tracking
Platform Capabilities
Everything your AI agents need to be smart, safe, and observable
Compiled Context
Turn large documents, datasets, and conversation history into compressed, reusable memory. Your AI remembers everything without hitting token limits. 85%+ compression with instant recall.
One-Click Integrations
Connect to Facebook, Google Ads, LinkedIn, TikTok, Postgres, and more with OAuth. No token management, no refresh logic—just plug and query. Your AI gets the data, you keep control.
Real-Time Observability
See every decision your AI makes as it happens. Track which context was loaded, what sources influenced the answer, and drill down into network calls. Replay any session instantly.
Policy & Risk Engine
Define guardrails: budget limits, allowed actions, approval workflows. Every request is scored for risk and checked against your policies before execution. Block bad actors automatically.
MCP Server for AI Agents
Connect Claude, Cursor, Replit, or any MCP-compatible AI to marketing APIs instantly
MCP Server URL
Add this to your MCP client configuration
https://datagran-mcp.fly.dev/mcpClaude Desktop
Add to your claude_desktop_config.json:
{
"mcpServers": {
"datagran": {
"url": "https://datagran-mcp.fly.dev/mcp"
}
}
}Cursor IDE
Add to your .cursor/mcp.json:
{
"mcpServers": {
"datagran": {
"url": "https://datagran-mcp.fly.dev/mcp"
}
}
}Available MCP Tools
Brain & Memory
query_brain
Query unified memory (recommended first step)
compile_context
Manually compile data to memory
query_with_context
Query specific memory by ID
list_compiled_memories
List all compiled memories
Connection & Authentication
check_connection_status
Check if a provider is connected
show_marketing_console
Open OAuth widget to connect
Data Fetching
execute_script
Run API calls on marketing platforms
postgres_query
Query PostgreSQL databases
google_ads_accounts
List Google Ads accounts
Web Pixel Analytics
web_pixel_get_stats
Get analytics: visitors, sessions, signups
web_pixel_list_users
List identified users with emails
web_pixel_list_sites
List all pixel sites
Web Scraping (Firecrawl)
firecrawl_scrape
Scrape single URL content
firecrawl_crawl
Crawl entire websites
firecrawl_map
Map all URLs on a website
firecrawl_search
Web search with scraping
firecrawl_extract
Extract structured data with AI
Supports Facebook Ads • Instagram • Google Ads • Google Drive • Google Calendar • Salesforce • LinkedIn Ads • TikTok Ads • Postgres • Firecrawl • Web Pixel
Knowledge Sources
One-click access to the platforms your AI needs
Connect in One Click
Facebook Ads
Access ad accounts, campaigns, ad sets, and performance metrics.
Facebook Leads
Capture lead submissions via the dedicated leads_retrieval scope.
Read account insights via the instagram_manage_insights scope.
Google Ads
Auto-link to our manager, execute scripts, and centralize spend + performance.
LinkedIn Ads (Closed Beta)
Map Sponsored Accounts once and launch campaigns or creatives via our proxy. Scopes: r_ads_reporting · rw_organization_admin · r_ads · rw_ads.
Google Drive
Read files, metadata, and exports via our secure proxy.
Google Calendar
Create events, check availability, and manage calendars via our secure proxy.
Gmail
Read emails, search inbox, and send messages via our secure proxy.
Salesforce
Query and write CRM data via the Salesforce REST API—securely proxied by Datagran.
TikTok Ads
Full marketing API access—advertisers, campaigns, reporting, leads, measurement, and ad comments.
Postgres / Supabase
Full SQL proxy with encrypted credentials, SSRF protection, audit logs, and query observability.
Firecrawl
Web scraping, crawling, and structured data extraction. Scrape any URL, crawl entire sites, or extract data using LLM schemas.
Web Pixel
First-party analytics pixel. Track page views, sign-ups, sign-ins, and identify users with email. Lightweight JS snippet.
Bank-Level Security
Your users trust you with their data. We protect it like our own.
AES-256-GCM Encryption
All OAuth tokens are encrypted at rest using military-grade AES-256-GCM encryption. Each token has its own unique initialization vector and authentication tag for maximum security.
Zero Token Exposure
Tokens never touch your application or frontend. All API requests go through our secure data proxy—your partners only receive the data they need, never the raw credentials.
PKCE OAuth Flow
We implement OAuth 2.0 with PKCE (Proof Key for Code Exchange) to prevent authorization code interception attacks. Every auth session has unique challenge/verifier pairs.
API Key Hashing
Partner API keys are hashed with bcrypt (10 rounds) before storage. We never store plain-text keys—even we can't see your partners' credentials after creation.
Origin Whitelisting
Strict CORS enforcement and origin validation. The auth widget only opens from pre-approved domains, preventing unauthorized access attempts and phishing attacks.
Audit Logs
Complete audit trail of all authentication events, token refreshes, and API access. Track every action for compliance and security monitoring.
Data Deletion & Compliance
Full GDPR compliance with automated data deletion workflows. When users disconnect or delete their accounts, we automatically revoke tokens and purge all related data from our systems.
Ready to Supercharge Your Agents?
Give your AI the context it needs, the knowledge it deserves, and the observability you require. All in minutes.