Description
< #AI Phone Assistant #Customer Support #Self-Hosted #SIP #Omnichannel #AI Agent #Open Source
Tel-Agent is an open-source phone gateway with a very specific place in the stack: it sits between a phone line and an AI agent. A call arrives over SIP, gets checked against your routing rules, and is then passed to a human, blocked, or handed to the AI. The agent talks to the caller in real time and can invoke tools along the way — transfer the call, take a message, check a calendar, hit any HTTP endpoint — while every call is recorded, transcribed and made searchable. All of it runs on your own hardware with your own API keys.
One thing to be clear about up front: the project is pre-alpha and there is no installable release yet. It is at milestone 0 of 12, and the phone — the hardest path — comes last; the author is getting one conversation working end to end in web chat first, then the messaging channels. So right now it suits people who want to follow the progress, contribute, or evaluate a self-hosted approach early, rather than anyone who needs to ship today. It is maintained by Dpro GmbH in Vienna under AGPL-3.0.
Plugs into your existing PBX: It registers as an extension on 3CX, Asterisk or FreePBX — it is not a PBX replacement and not a workflow platform. A genuinely analog line needs an ATA to bridge it to SIP.
Caller routing rules: Every call passes through your rules before anything else, so you decide which callers reach a human, get blocked, or ever talk to the AI at all.
Real-time conversation with tools: During a call the agent can transfer, take messages, check a calendar, or call any HTTP endpoint, which is how systems like n8n and Home Assistant get reached.
Recordings, transcripts and search: Every conversation is recorded, transcribed and indexed, and the recordings stay on your own machine rather than on a third-party platform.
Ten channels, one agent: Phone, web chat, SMS, email, WhatsApp, Telegram, Messenger, Instagram, Discord and Slack — each connected with your own credentials from that platform.
Bring your own model keys: Models are connected over MCP, covering OpenAI, Claude, Gemini, Mistral, DeepSeek, Qwen, Ollama and more, so switching models does not mean rebuilding the setup.
Self-hosted by design: Docker Compose brings the stack up, both ports bind to 127.0.0.1 by default, the first visit forces an administrator account with no default credentials, API keys are encrypted at rest, and recording announcements are on by default.
Planned architecture: LiveKit Agents for voice, FastAPI for the API, Next.js for the frontend, PostgreSQL for transcripts and Redis for queues, with a target of under 800 ms from end of speech to first audio out.
Open source project: The source code is hosted on GitHub, so developers can study the implementation, contribute, or customize it for their own needs.
Tel-Agent is an open-source phone gateway with a very specific place in the stack: it sits between a phone line and an AI agent. A call arrives over SIP, gets checked against your routing rules, and is then passed to a human, blocked, or handed to the AI. The agent talks to the caller in real time and can invoke tools along the way — transfer the call, take a message, check a calendar, hit any HTTP endpoint — while every call is recorded, transcribed and made searchable. All of it runs on your own hardware with your own API keys.
One thing to be clear about up front: the project is pre-alpha and there is no installable release yet. It is at milestone 0 of 12, and the phone — the hardest path — comes last; the author is getting one conversation working end to end in web chat first, then the messaging channels. So right now it suits people who want to follow the progress, contribute, or evaluate a self-hosted approach early, rather than anyone who needs to ship today. It is maintained by Dpro GmbH in Vienna under AGPL-3.0.
Features
Plugs into your existing PBX: It registers as an extension on 3CX, Asterisk or FreePBX — it is not a PBX replacement and not a workflow platform. A genuinely analog line needs an ATA to bridge it to SIP.
Caller routing rules: Every call passes through your rules before anything else, so you decide which callers reach a human, get blocked, or ever talk to the AI at all.
Real-time conversation with tools: During a call the agent can transfer, take messages, check a calendar, or call any HTTP endpoint, which is how systems like n8n and Home Assistant get reached.
Recordings, transcripts and search: Every conversation is recorded, transcribed and indexed, and the recordings stay on your own machine rather than on a third-party platform.
Ten channels, one agent: Phone, web chat, SMS, email, WhatsApp, Telegram, Messenger, Instagram, Discord and Slack — each connected with your own credentials from that platform.
Bring your own model keys: Models are connected over MCP, covering OpenAI, Claude, Gemini, Mistral, DeepSeek, Qwen, Ollama and more, so switching models does not mean rebuilding the setup.
Self-hosted by design: Docker Compose brings the stack up, both ports bind to 127.0.0.1 by default, the first visit forces an administrator account with no default credentials, API keys are encrypted at rest, and recording announcements are on by default.
Planned architecture: LiveKit Agents for voice, FastAPI for the API, Next.js for the frontend, PostgreSQL for transcripts and Redis for queues, with a target of under 800 ms from end of speech to first audio out.
Open source project: The source code is hosted on GitHub, so developers can study the implementation, contribute, or customize it for their own needs.
