How swarmates.ai works

Every agent.One swarm.

Your agents live in 10 different tabs today. swarmates.ai connects them into one coordinated swarm: one vault for their keys, one library for shared context, one fabric for automations, and one audit log for every message passing between them. Abacus AI, ChatGPT, Claude, or your own custom agent. All moving as one.

The four building blocks

Everything you do in the swarm comes down to combining these four things.

Agents

Any AI service, anywhere.

Abacus, OpenAI, Claude, or any REST webhook.
Credentials encrypted with AES-256-GCM at rest.
Auto-pinged for health, so you see status at a glance.

Context

The shared brain your agents read from.

Upload files, paste schemas, drop endpoints and notes.
Pick exactly which agents can see which items.
Stored in private S3 with presigned access only.

Connections and messages

Permission and an audit trail, in one click.

A connection says: Agent A may call Agent B.
Every call is a message: payload in, response out, tokens, latency.
The activity feed is live, via Server Sent Events.

Automations

When this, do that. Zero code.

Triggers: manual, schedule, agent event, webhook.
Actions: query another agent, with chosen context.
Dry-run mode to preview without real calls.

The five-minute story

From signup to live automation, the path looks like this.

1STEP

Sign up

Free plan, no card. Two agents and unlimited messages between them forever.

2STEP

Add agents

Wizard guides you: pick type, paste key, test connection live.

3STEP

Share context

Drop files or notes into the library. Attach them to the agents that need them.

4STEP

Build automation

When Agent A says X, send it to Agent B with context. Then, maybe, to Slack.

5STEP

Flip it live

Dry-run once, then enable. The activity feed shows every call in real time.

Under the hood

swarmates.ai is built on a small set of boring, proven pieces, wired together with care.

AES-256-GCM at rest
API keys are encrypted with a key derived from your secret. The browser never sees a raw key again.
Server-Sent Events
Real-time activity pushed to your browser. No polling, no WebSocket fragility.
Adapter pattern
Each provider (Abacus, OpenAI, Claude, webhook) implements the same three functions. Adding Gemini is 80 lines.
Per-user isolation
Every query is scoped by user id at the database layer. No cross-tenant leakage.

LIVE DATA FLOW

user.ticket→Agent A (GPT-4o)classify

Agent A→Agent B (Claude)+ context.pdf

Agent B→webhook.helpdeskcreate ticket

result→activity feedlogged, audited

Every hop is logged. Every payload is encrypted. Every key stays on the server.

Voice dictation is on

Listening in your selected language…

“When Agent A finishes a draft, hand it off to Agent B for proofreading, then post the result to our Slack channel.”

en-USes-ESfr-FRde-DEzh-CNja-JPpt-BR

New: voice prompting

Talk to it. It writes it down.

Every text field that matters, agent descriptions, system prompts, context notes, automation templates, messages, now has a microphone button. Tap it, speak, and watch your words land in the field.

Runs on your device using the browser's built-in speech recognition.
Audio never leaves your machine. Zero added cost.
Automatically follows your UI language. Seven supported out of the box.
Works in Chrome, Edge, Safari on desktop and mobile.

What you can build in an hour

Three concrete examples with two or three agents each.

MARKETING

Draft to CMS pipeline

1Agent A (GPT-4o) writes a blog draft.
2Agent B (Claude) proofreads and flags facts.
3Webhook publishes the approved draft.

SUPPORT

Auto ticket triage

1Your chat agent collects a ticket.
2An OpenAI agent classifies it as bug, billing, or how-to.
3Webhook creates the ticket in Zendesk.

DATA

Dataset copilot

1A dataset lands in the Context library.
2Claude summarises the schema.
3A code agent writes the ETL, opens a PR on GitHub.

Ready to wire up your agents?

Two agents are free forever. No credit card. You can be live in five minutes.