Platform Overview
Ajeris is a personal AI agent platform. Each user gets their own agent, with its own memory and connections kept apart from every other account's, that handles email, calendar, ride-booking, smart home, music, shopping, web search, messaging, and any custom tool the user connects, all from a text message or a voice command.
The core design principle is one agent, many surfaces. You talk to the same agent whether you're speaking to Alexa, texting via iMessage, or using a future web or mobile client. Conversation is continuous across surfaces: a voice turn can reference an SMS from two minutes ago, and the agent can push a tappable link to your phone while it's still speaking.
How a turn works
When you message your agent, the same broad pipeline runs whether the message arrived as voice or as text:
- Receive. Twilio (for SMS) or Amazon (for Alexa) hands the message to Ajeris.
- Identify. Ajeris resolves which user is sending the message
- by phone number for SMS, by linked-account identity for voice.
- Load context. The agent loads your relevant memories, the tail of your recent conversation history (across all surfaces), and the tools you've authorized.
- Think. The agent decides what to do, calling whichever tools it needs (read your calendar, send an email, place an order, look something up online, etc.).
- Reply. The agent composes a reply tuned to the surface you're on - SMS replies can include emoji, links, and longer text; voice replies are formatted for spoken delivery and avoid reading URLs aloud (those get pushed to your phone as an SMS link).
- Log. The exchange is appended to your conversation history so future turns have it as context.
A typical turn finishes in well under a second of overhead before the model starts generating. Tool calls add the latency of whichever external API the tool touches.
Surfaces
| Surface | Description |
|---|---|
| SMS / iMessage | Text your Ajeris number from anywhere. Default surface for most users. Replies can include links, emoji, and longer text. |
| Voice (Alexa) | "Hey [your agent], …" on any Alexa device. Replies are spoken; long text and links get sent to your phone as SMS. |
| Web / mobile | Coming soon. Same agent, same memory, same tools - just a different way to reach it. |
The agent knows which surface a turn arrived on and adapts its reply. URLs in a voice answer are automatically delivered as SMS links; long lists are summarized aloud with the full list pushed to your phone.
Memory and continuity
Ajeris is meant to feel like an assistant who knows you, not a tool that starts cold every time. Two mechanisms make that possible:
Conversation history. A windowed slice of your recent turns (across all surfaces) is included in every request. That means a voice question can pick up an SMS thread from a few minutes earlier.
Long-term memory. Facts, preferences, and routines worth keeping are saved into a per-user memory store. Some are saved explicitly when you ask ("remember that…"); others are picked up passively from conversation. See Memory model for the full breakdown.
Tools
Tools are the things your agent can actually do. Two tiers:
Built-in integrations
These are first-party integrations Ajeris ships with. You authorize them once during onboarding and the agent can use them from then on.
| Service | Examples of what you can ask |
|---|---|
| Gmail | Check unread, summarize, draft, send, reply, search |
| Google Calendar | What's today, what's next, schedule, cancel, find free time |
| Google Drive | Find docs, summarize, share |
| Spotify | Play, pause, queue, switch device |
| Apple Music | Play, search, manage playlists |
| YouTube | Search, summarize a video, queue |
| Hue | Lights on/off, scenes, brightness, by room |
| Uber / Lyft | Request a ride to a saved place via deep link |
| DoorDash | Reorder usuals, search restaurants via deep link |
| Slack | Read channels, send messages, search, react |
| GitHub | List PRs, issues, commits; comment; review |
| Notion | Search, read, create pages, append to databases |
| HubSpot | Read contacts, companies, deals; (more coming) |
| Shopify | Orders, products, refunds, gift cards (per shop) |
| QuickBooks / Xero | Read accounting reports, contacts, invoices |
| Stripe / Square | Read payments, balances, recent transactions |
| Plaid | Bank balances and recent transactions (planned) |
| Web search / fetch | Live web search and page reading |
This list grows over time. Anything not in the list can usually be handled by connecting your own MCP server (next section).
Bring-your-own MCP
If you operate a service that exposes an MCP server, you can connect it to Ajeris and the agent will gain that server's tools. This is how power users plug in their own internal company APIs, niche services, or community MCP servers.
See Connect your MCP server for the full integration guide.
Security and privacy
The full details live in the security policy and the privacy policy, which has the full retention schedule. At a high level:
- Per-user isolation. Your data is strictly scoped to your account at the database layer. One user's agent cannot see another user's data.
- Encryption at rest. OAuth tokens and custom-MCP credentials are encrypted before they're stored.
- Staff access. Nobody at Ajeris reads your data except when you ask, for security or legal reasons, or as the privacy policy describes for improving Ajeris.
- You delete it, it's gone. Removing a connected service invalidates the stored credential. Removing your account purges your memories, conversations, and tokens from our database; the privacy policy lists what providers keep afterwards, and for how long.
Tech stack
For developers and integrators curious about the underlying choices:
| Layer | Technology |
|---|---|
| Runtime | Node.js + TypeScript |
| AI | Claude Agent SDK with multiple Claude model tiers |
| Tools | Model Context Protocol (Streamable HTTP and stdio) |
| Database | PostgreSQL with row-level isolation and encryption at rest |
| SMS | Twilio Messaging Service |
| Voice | Alexa Skills Kit |
| Hosting | One shared agent service; each account's data separated by row-level security |
The agent uses the Model Context Protocol for all of its tools. That's how built-in integrations and bring-your-own servers can sit side-by-side without the agent caring which is which.