Self-hosted AI agents
that work where you do.

Talon is an open-source runtime for useful, long-lived AI workflows. Connect the channels and tools you already use, choose the model endpoint, and keep control of what each persona can do.

View the source on GitHub →

Useful work, in the places you already use

These are configuration patterns, not boxed products. You choose the integrations, permissions, and review points.

Your day, prepared

Get a morning brief, meeting context, reminders, and follow-ups in the channel you actually check. Connect the calendar and work tools you choose.

Household coordination, in one thread

Capture shopping, delivery, and home requests as they arise. Add trusted MCP integrations and set the boundaries before anything can act.

Small-team operations without another dashboard

Route requests, prepare summaries, and run scheduled check-ins across the tools your team already uses. Give each persona only the access it needs.

Control the parts that matter

Runs where you choose

Deploy on infrastructure you control. You choose the model providers and integrations Talon connects to, then assess their data handling before sending them context.

Keeps work moving

Durable queueing, threads, schedules, and background work let an assistant continue a workflow beyond one chat response.

Makes boundaries explicit

Personas receive only the configured host-tool capabilities. Talon records side-effecting host-tool operations so you can inspect what happened.

How it works

The model call is one piece. Talon handles the surrounding work.

1

Connect a channel

Start with the channel people already use: Telegram, Slack, WhatsApp, terminal, Discord, or email.

2

Define a persona

Give it a purpose, model provider, skills, and only the host tools it needs for that job.

3

Add the workflow

Connect trusted integrations and schedules. Talon routes, persists, and runs the work on your infrastructure.

Connect the parts you already use

Choose a channel, then use the model runtime and MCP integrations that fit the deployment.

Telegram Slack WhatsApp Terminal Discord Email

Talon keeps personas, channels, scheduling, and host-tool policy in its own layer while provider adapters execute the model call. See the provider reference →

Run your first useful workflow

Start with one persona and one channel. Talon is open source under AGPL-3.0; you bring the infrastructure and model provider that fit your needs.

# A small first deployment:
1. Pick a channel
2. Configure a provider and persona
3. Give it only the capabilities it needs
Then add memory, schedules, and MCPs when the workflow calls for them.

Project updates

Recent activity from ivo-toby/talon.