Backside is a cloud API and MCP server that gives AI agents structured, encrypted access to contacts, CRM, tasks, calendar, and notes — and a durable executor that can run agents against that data for you. Connect once. Every agent you run gets the same data layer.
Coming soon.
Backside is a headless personal information manager — a cloud API and MCP server that stores and serves structured personal data for AI agents and developers. No product UI. No paste-your-context-again ritual.
It covers five data domains: Contacts, CRM, Tasks & Projects, Calendar, and Notes. Data is encrypted at rest with AES-256, encrypted in transit with TLS 1.3, and PII fields are encrypted at the column level before they ever reach the database.
Backside implements the vCard and iCalendar open standards, so everything is portable and exportable. Developers access data via a REST API or MCP server over Streamable HTTP. AI agents can pay per request in USDC over the x402 protocol — no subscription, no API key management.
Backside also runs agents for you. Register a config, POST to an endpoint, and a worker picks up the run, drives the LLM loop, calls your data, and journals every step. Runs survive worker restarts, enforce guardrails on every tool call, and use your own Anthropic key for LLM calls.
Add Backside to any AI client that speaks the Model Context Protocol. Your agent gets structured access to all five data domains immediately.
{
"mcpServers": {
"backside": {
"url": "https://mcp.backside.app/mcp",
"headers": {
"Authorization": "Bearer bsk_live_your_key_here"
}
}
}
}Or use the REST API directly. Standard HTTPS. JSON in, JSON out. Works with any language, any framework, any agent runtime.
Every domain is modeled after an open standard wherever one exists. No proprietary schemas to reverse-engineer.
| Domain | Description |
|---|---|
| Contacts | People, organizations, relationships, interaction history, communication channels |
| CRM | Pipelines, deals, stages, revenue tracking, deal-to-contact role mapping |
| Tasks & Projects | Task management, subtasks, dependencies, milestones, RRULE recurrence |
| Calendar | Events, scheduling, availability, attendees, all-day and timed events |
| Notes | Structured and unstructured notes, notebooks, tags, bidirectional wikilinks |
Three layers of encryption. No fallback to weaker modes. The database only ever sees ciphertext for the fields that matter.
Every byte on disk is encrypted. Database volumes, backups, and WAL archives — no plaintext ever touches the filesystem.
No fallback to older protocols. No weak ciphers. Cloudflare origin certs plus mTLS authenticated origin pulls between edge and origin.
PII fields are encrypted at the application layer with AES-256-GCM before they reach the database. The DB only ever sees ciphertext.
A single URL your agent talks to over Streamable HTTP. No local process to babysit. Your agent only sees the domains you enable.
Remote server, not a local process. Connect from Claude Desktop, Cursor, or any MCP client that speaks Streamable HTTP.
Your agent only sees the domains you enable. No tool sprawl, no cluttered prompts, no accidental access to things you didn't opt into.
Create contacts, schedule events, manage tasks, log interactions — all through natural language. Every domain is first-class.
Ask "what's on my plate this week?" and get tasks, calendar events, and CRM deals in one response — not three round trips.
Don't want to stand up your own runner? Register a config, POST to
/api/v1/agents/{kind}/runs, and a worker drives
the LLM ↔ tool loop against your data. The HTTP call returns
immediately — runs go async. Every step is journaled.
Backside never pays for your LLM tokens. Keys are encrypted per tenant with AES-256-GCM and loaded only at the worker boundary. Rotate or revoke any time.
Allowlist, denylist, rate limit, approval gate, IO validation, quiet hours — with shadow and enforce modes. A brand-new config defaults to deny, so agents cannot accidentally mutate data on day one.
Per-run budget in USD cents, per-kind auto-disable after ten failed runs in 48 hours, daily rollups in a billing view. You can see every cent.
Content-hashed journal in Postgres with bit-identical replay, 30-second leases with heartbeat, orphan recovery on worker crashes, OpenTelemetry GenAI semconv spans on every LLM and tool call.
Humans want predictable monthly billing. Agents want permissionless micropayments. Backside serves both without forcing either.
Monthly Stripe subscription. API key provisioning, rate limits, and a dashboard for usage and billing. Traditional billing for people who want predictability.
Per-request USDC payments on Base via the x402 protocol. No account, no API key, no KYC. The agent includes payment in the HTTP header and gets immediate access. Permissionless.
Whether you're using AI every day or shipping a product on top of it, Backside is the data layer you don't have to design yourself.
You use Claude Desktop or Cursor every day. You want your agent to know who you know, what's on your calendar, and what you're working on — without pasting context into every prompt. Backside gives your agent a memory that works across every AI provider. Switch models without losing your data.
You're building an AI assistant, a smart CRM, or an agent-powered scheduling tool. You don't want to spend months designing a contacts-tasks-calendar schema from scratch. Backside gives you 20+ tables, multi-tenant isolation, and encryption out of the box. Focus on what differentiates your product.