---
title: "Tel-Agent: Is This Open-Source AI Phone Assistant Really Self-Hosted?"
date: "2026-09-14"
video_id: "7685260288450448660"
tiktok_url: "https://www.tiktok.com/@github.signals/video/7685260288450448660"
creator: "@github.signals"
category: "AI"
tags: ["tel-agent", "ai phone assistant", "livekit", "self-hosted ai", "agpl"]
excerpt: "Tel-Agent connects a phone line to an AI agent. What ships today, where call audio goes by default, why its award badge is another product's and how it is licensed."
---
Tel-Agent is an AGPL-3.0 gateway from Vienna's Dpro GmbH that sits between a phone line and an AI agent, routing calls and messages to a model you choose. A TikTok from @github.signals presents it as an AI phone assistant that "runs entirely on your own hardware." The repository is real, but its default setup sends call audio to LiveKit Cloud, Deepgram, OpenAI and ElevenLabs, the phone feature is milestone 11 of 12 on its roadmap, and its own docs say the "Product of the day" badge belongs to a different product.

## What is Tel-Agent?

The [README](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/README.md) says: "Tel-Agent is an open-source gateway that sits between a phone line and an AI agent. A call arrives over SIP. Tel-Agent checks the caller against your routing rules and either passes it through to a human, blocks it, or hands it to an AI agent." Its tagline is "Connect any phone line to any AI model. Self-hosted, bring your own keys."

It also lists what it is not: a PBX replacement, a workflow automation platform, a CRM or anything that works with analog lines.

The [repository](https://github.com/Dpro-at/Tel-Agent) was created on 2026-08-16. When checked on 2026-09-15 it had 516 stars, 115 forks and 210 commits, with releases v0.1.0 and v0.1.1 both on 2026-09-05. The stack is LiveKit Agents, FastAPI, Next.js, SQLite or Postgres, Redis and Caddy. Of the 210 commits, 148 come from one account that commits as Mohamed Alarade, Dpro's managing director. The README carries a "Built with Antigravity and Codex" badge.

## Does Tel-Agent run entirely on your own hardware?

Not in its default configuration. The [`.env.example`](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/.env.example) sets `TELEPHONY_MODE=livekit`, described as "A number bought from a SIP provider, pointed at a LiveKit Cloud SIP trunk." An on-premises mode, `TELEPHONY_MODE=sip`, registers directly to a 3CX, Asterisk or FreePBX extension, but it is not the default.

The speech pipeline is hosted too. The defaults point to `api.deepgram.com` for speech-to-text, `api.openai.com` for the model and `api.elevenlabs.io` for the voice. The README says: "Providers for v1: Deepgram (STT) · one cloud LLM · ElevenLabs (TTS). Local models (Ollama, Whisper, Piper) follow in v1.1."

The project's own [blog post on GDPR](https://tel-agent.com/blog/gdpr-self-hosted-ai-phone-assistant) concedes this: "The design that gets the first call answered uses hosted speech-to-text, a hosted model and a hosted voice." Recordings and transcripts are stored on your machine. The caller's audio still leaves it.

## Are the security defaults good?

Yes, and the video skips them. The server binds to `127.0.0.1` by default, a password is required on first run with no default credentials, and API keys are encrypted at rest. Recording announcements are on by default (`RECORDING_ANNOUNCEMENT=true`).

The README justifies that last default with "Austria requires both parties to be aware that a call is recorded." Austria's [StGB § 120](https://www.ris.bka.gv.at/NormDokument.wxe?Abfrage=Bundesnormen&Gesetzesnummer=10002296&Paragraf=120) penalises unauthorised recording of non-public statements with up to one year in prison, and practitioner guidance reads it as requiring consent, which is stricter than awareness.

## Did Tel-Agent win Product of the Day?

No. The "Product of the day / 1st" seal on the README is a local SVG file. The repository's [awards README](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/docs/brand/awards/README.md) says it is "The wreath lockup from the Flowxtra registration page" and that "'Product of the day - 1st' is **Flowxtra's**, earned by a shipped product. Tel-Agent is pre-alpha, has no release, and has not placed anywhere." Flowxtra is a separate Dpro recruitment product. I found no Product Hunt listing for Tel-Agent.

The "DSGVO & GDPR COMPLIANT" and "AI ACT COMPLIANT" badges are also reused Flowxtra artwork. The [compliance README](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/docs/brand/compliance/README.md) says that "GDPR and the AI Act are, for a self-hosted product, properties of a deployment rather than of a repository."

Under [EU AI Act Article 50](https://artificialintelligenceact.eu/article/50/), applicable from 2 August 2026, people must be told when they are talking to an AI system unless it is obvious. That duty falls on whoever deploys the phone agent.

## Can Tel-Agent answer phone calls yet?

Not as a finished feature. The [roadmap](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/docs/ROADMAP.md) has twelve milestones and the rule "Step N+1 does not start before step N works." The project is at milestone 3, messaging channels.

Phone calls are milestone 11. The transport code exists, with streaming speech, interruption handling, caller-ID routing and stored transcripts. The roadmap says what is missing: "the latency budget, endpointing and barge-in on a live line." None of its six acceptance checks are ticked. The target is under 800 ms from the end of caller speech to first audio, tested on at least 20 real calls in Austrian German. It warns, "Text-channel comfort is not evidence."

## How many channels does Tel-Agent support?

The README says the "official channel scope now covers 24 channels including the phone." The wired channels are web chat, WhatsApp, Telegram, Messenger, Instagram, Discord, Slack, email and SMS. Microsoft Teams, Signal and Viber were merged on 2026-09-14. Seven channels, including iMessage and Google Chat, were still open issues on 2026-09-15. The 24 is a plan.

## Is Tel-Agent free for commercial use?

It is AGPL-3.0. The README says, "If you run a modified version of Tel-Agent as a network service, you must publish your modifications." Dpro sells a commercial license for closed-source integration.

Every contributor signs a [CLA](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/CLA.md) giving Dpro GmbH an "irrevocable" license that "includes the right to relicense Your Contribution, including under proprietary or commercial license terms," so it can offer "a hosted edition and a commercial license." This is a common dual-license model, and one company holds the commercial rights to all contributions.

## Bottom line

Tel-Agent is a real, fast-moving AGPL project with good security defaults and working text channels. Its default setup sends call audio to hosted providers, its phone feature has not passed its own tests, and its award and compliance badges are borrowed from another Dpro product, as its own docs admit.

## Sources

- [Dpro-at/Tel-Agent on GitHub](https://github.com/Dpro-at/Tel-Agent) - the repository
- [Tel-Agent README](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/README.md) - features, providers and security notes
- [Tel-Agent .env.example](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/.env.example) - default telephony and provider settings
- [Award seal documentation](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/docs/brand/awards/README.md) - origin of the Product of the day badge
- [Compliance badge documentation](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/docs/brand/compliance/README.md) - origin of the GDPR and AI Act badges
- [ROADMAP.md](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/docs/ROADMAP.md) - milestones and phone call criteria
- [CLA.md](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/CLA.md) - contributor licence terms
- [The GDPR case for a self-hosted AI phone assistant](https://tel-agent.com/blog/gdpr-self-hosted-ai-phone-assistant) - the project's own blog post
- [tel-agent.com](https://tel-agent.com) - project site
- [Dpro GmbH](https://dpro.at/en/about-us/) - the company and its products
- [livekit/agents](https://github.com/livekit/agents) - the voice agent framework Tel-Agent builds on
- [EU AI Act Article 50](https://artificialintelligenceact.eu/article/50/) - AI disclosure obligation
- [Austrian StGB § 120 on RIS](https://www.ris.bka.gv.at/NormDokument.wxe?Abfrage=Bundesnormen&Gesetzesnummer=10002296&Paragraf=120) - the recording statute
- [Open Source AI Phone Assistant on DEV](https://dev.to/9mtm/open-source-ai-phone-assistant-call-chat-25oe) - the maintainer's writeup


## Video description and verification notes

**The repository in this video keeps a file explaining that the "Product of the day, 1st" seal on its README was taken from a different product's marketing page and that Tel-Agent "has not placed anywhere," and the roadmap in the same repository puts the phone call feature the video leads with at milestone 11 of 12 while the project sits at milestone 3.** The clip is 38 seconds, 720x1280, a synthetic voiceover over a slow screen recording of `github.com/dpro-at/tel-agent` in Chrome, with word-by-word yellow-highlight captions burned in and a `@GithubSignals` watermark. Nothing is demonstrated. The README scrolls from the header to the "Why it exists" section and then cuts to a branded outro card reading "Follow for more open-source projects." When I pulled the metadata on 2026-09-15 it showed 21,800 views, 1,086 likes, 1,480 saves, 260 reposts, and 7 comments.

## What the 38 seconds shows

The video is a single scroll. Frame one opens mid-page with the tab title "GitHub - Dpro-at/Tel-Agent" and the README tabs visible: README, Contributing, AGPL-3.0 license, Security. Below the dove-and-bullet logo the header reads "AI phone assistant" and then "Connect any phone line to any AI model. Self-hosted, bring your own keys." Two rows of channel icons follow, then the line "Connect them over MCP, and get full control," then a row of thirteen model logos.

Under those sit the three things worth stopping on. A laurel wreath seal reading "Product of the day / 1st." Two badges reading "DSGVO & GDPR COMPLIANT" and "AI ACT COMPLIANT." Then three shields: "License AGPL v3," "status alpha," "release v0.1.1." The attribution line reads "tel-agent.com · maintained by Dpro GmbH, Vienna."

The body text the scroll passes over is legible in frames 8 through 18. "Tel-Agent is an open-source gateway that sits between a phone line and an AI agent. A call arrives over SIP. Tel-Agent checks the caller against your routing rules and either passes it through to a human, blocks it, or hands it to an AI agent." Then "It runs on your own hardware, on your own LAN, with your own API keys." A "What it is not" list follows: not a PBX replacement, not a workflow automation platform, not a CRM, not analog-capable. The scroll stops at "Why it exists," which reads "Well-funded closed products already do 'AI receptionist for business'."

The voiceover is a near-verbatim reading of the README's first three paragraphs plus the TikTok description. The captions transcribe the product name as "TELAGENT."

## The repository is real, four weeks old, and built at unusual speed

I pulled the GitHub API directly. The repository is `Dpro-at/Tel-Agent`, created 2026-08-16, last pushed 2026-09-14, licensed AGPL-3.0, 516 stars, 115 forks, 30 open issues, 210 commits on `main`. Language breakdown is roughly 2.4 MB of Python and 1.0 MB of TypeScript, with smaller amounts of CSS, Shell, PowerShell, Inno Setup, Dockerfile, and Batchfile. Two releases exist: `v0.1.0` marked prerelease on 2026-09-05 at 12:52 UTC, and `v0.1.1` four hours later at 16:01 UTC with 7 attached assets.

The contributor distribution is lopsided. Ten contributors are listed. The top account, `9mtm`, has 148 commits. The second has 36, the third 16, the fourth 4, and the remaining six have one commit each. GitHub's participation stats show the weekly commit counts for the last ten weeks as 0, 0, 0, 0, 0, 6, 14, 84, 86, 20. Two consecutive weeks of 84 and 86 commits from a project with one primary author is fast.

The README explains part of that. It carries a badge reading "Built with Antigravity and Codex," and the repository root contains `CLAUDE.md`, `AGENTS.md`, `GEMINI.md`, a `.claude/` directory, and a `.cursor/` directory. This is an AI-agent-assisted codebase, which the project states openly rather than hiding.

The owner org is `Dpro-at`, display name "Dpro GmbH," Austria, created 2024-06-07, 7 public repos, 4 followers. The CLA file gives the full legal identity: Dpro GmbH, Wipplingerstraße 20/18, 1010 Wien, commercial register FN 631492s at Handelsgericht Wien, VAT ID ATU81090445, Managing Director Mohamed Alarade. The top contributor account `9mtm` commits under the name Mohamed Alarade, so the managing director is also the primary developer.

## "Runs entirely on your own hardware" is the claim the project's own blog walks back

The voiceover says "because it runs entirely on your own hardware, you keep full control over your data, your recordings, and your privacy." This is the load-bearing claim of the video and it does not survive contact with the configuration file.

I read `.env.example` from `main`. The default telephony mode is `TELEPHONY_MODE=livekit`, and the comment above it describes that mode as "A number bought from a SIP provider, pointed at a LiveKit Cloud SIP trunk." The variables under it are `LIVEKIT_URL`, `LIVEKIT_API_KEY`, `LIVEKIT_API_SECRET`, and `LIVEKIT_SIP_TRUNK_ID`, annotated "From the LiveKit Cloud project settings." The on-premises path exists as a second mode, `TELEPHONY_MODE=sip`, which registers directly to a 3CX, Asterisk, or FreePBX extension. It is not the default.

The speech pipeline is worse for the privacy claim. The defaults in the same file are `DEEPGRAM_BASE_URL=wss://api.deepgram.com`, `LLM_BASE_URL=https://api.openai.com/v1`, and `ELEVENLABS_BASE_URL=https://api.elevenlabs.io`. The README states it plainly in the architecture table: "Providers for v1: Deepgram (STT) · one cloud LLM · ElevenLabs (TTS). Local models (Ollama, Whisper, Piper) follow in v1.1." So in the shipped version, the caller's audio goes to Deepgram, the transcript goes to a cloud LLM, and the synthesized reply comes back from ElevenLabs. What stays on your hardware is the stored recording and the transcript archive, which is real and worth something, but it is not the same as the audio never leaving.

Tel-Agent's own blog agrees with me. The post at `tel-agent.com/blog/gdpr-self-hosted-ai-phone-assistant` argues that hosted deployments mean "the audio leaves the EU, and Chapter V applies: standard contractual clauses, a transfer impact assessment, and a dependence on the EU-US Data Privacy Framework." It then concedes the current state directly: "The design that gets the first call answered uses hosted speech-to-text, a hosted model and a hosted voice." The video took the project's aspiration and stated it as the present tense.

The one accurate privacy detail the video does not mention is better than the one it does. The README's security notes say the server binds to `127.0.0.1` by default, a password is required on first run with no default credentials, and API keys are encrypted at rest. I confirmed `BIND_HOST=127.0.0.1` and `BIND_PORT=38472` in the env file. Those are sensible defaults and they are the actual security story here.

## The award seal and the compliance badges came from a different Dpro product

This is the part the video scrolls straight past, and it is documented in the repository itself.

The "Product of the day / 1st" wreath is not a Product Hunt embed. It is two local SVG files, `product-of-the-day-light.svg` and `product-of-the-day-dark.svg`, sitting in `docs/brand/awards/`. That directory has a `README.md`, and I read it. It opens: "The wreath lockup from the Flowxtra registration page." Further down, under a heading titled "What this seal claims," it says: "'Product of the day - 1st' is **Flowxtra's**, earned by a shipped product. Tel-Agent is pre-alpha, has no release, and has not placed anywhere." It adds that "the seal transfers one product's standing onto another" and calls it "Worth a decision before the repository gets attention."

Flowxtra is a separate Dpro product, an AI recruitment and CV tool, listed on Dpro's own portfolio page alongside the "KI-Telefonassistent." I searched Product Hunt for Tel-Agent and found no listing. I also could not confirm a Product Hunt "Product of the Day" placement for Flowxtra from public sources, so I cannot verify the original award either, only that the repository says it belongs to a different product.

The two compliance badges have the same origin. `docs/brand/compliance/README.md` opens with "Dpro GmbH badges, reused from the Flowxtra registration page." The same file documents three badges that were removed, and the reasoning is honest enough to be quoted: `aes-256.png` read "AES verified 256," which the author notes "implies an external audit of a crypto implementation, and Tel-Agent is pre-alpha with no release." The two that stayed get a caveat in the same file: "GDPR and the AI Act are, for a self-hosted product, properties of a deployment rather than of a repository," and "Worth keeping in mind before the badges are read as a certification."

One fairness note. The awards README says "pre-alpha, has no release," which was true when it was written and is now stale, because `v0.1.1` shipped 2026-09-05 and the main README says alpha. The badge is still in the README as of my fetch on 2026-09-15. The decision the file asks for does not appear to have been made.

For what it is worth, EU AI Act Article 50 became applicable on 2 August 2026 and requires that a person interacting with an AI system be told so unless it is obvious, with penalties up to 15 million euro or 3 percent of worldwide turnover. That obligation lands on whoever deploys the phone agent, not on the repository. A badge on a README does nothing for it.

## The phone is milestone 11 of 12, and the video opens with it

`docs/ROADMAP.md` is 18 KB and states the rule up front: "Twelve milestones. Step N+1 does not start before step N works." The status line reads "Milestone 3, messaging channels. Milestones 0 to 2 are closed, and the packaging of Milestone 10 shipped early."

Milestone 11 is "Phone call." Its own status paragraph says the transport exists ahead of its milestone, with streaming speech recognition and synthesis, interruption handling, caller-ID routing, and stored transcripts over SIP and LiveKit. Then it says what is still owed: "the latency budget, endpointing and barge-in on a live line, and no check here is ticked until it is taken on one." Six checks are listed (Arrive, Answer, Speak, Listen, Full loop, Same archive) and none of them are marked done.

The measurement criteria are specific and unmet. The target is under 800 ms from the end of caller speech to first audio out, plus "accuracy across at least 20 real calls in Austrian German including names and addresses," with a hard stop: "If latency exceeds ~1.5 s, nothing else is added until streaming is fixed." The roadmap also warns against exactly the inference this video makes: "Text-channel comfort is not evidence. No milestone before 11 may claim the latency question is answered."

The channel count deserves the same treatment. The README says "The official channel scope now covers 24 channels including the phone." Scope is the operative word. What is wired today, per both the README and the roadmap, is web chat, WhatsApp, Telegram, Messenger, Instagram, Discord, Slack, email, and SMS. The rest are open issues under the `channel` label. When I checked on 2026-09-15 there were 7 still open: iMessage, Feishu/Lark, DingTalk, QQ Bot, WeCom, WeChat Official Account, and Google Chat. Microsoft Teams, Signal, and Viber were merged on 2026-09-14, the same day the video posted. The video's "it also handles texts, emails and chat messages through the same brain" is accurate for the shipped channels. The "24" on screen is a plan.

The architecture underneath is credible. LiveKit Agents, the voice framework it builds on, is a real Apache 2.0 project with 14,205 stars and 3,739 forks, created 2023-10-19, and it is the standard choice for this job. FastAPI, Next.js, SQLite or Postgres, Redis, and Caddy are unremarkable in the good sense.

## AGPL-3.0 plus a relicensing CLA

The license shield is accurate and the implications go further than the video suggests. The README's license section says: "If you run a modified version of Tel-Agent as a network service, you must publish your modifications. The free version is never crippled; it is the product." It then offers "a commercial license permitting closed-source integration" via `info@dpro.at`.

That is a dual-license business model, and the CLA is the mechanism. `CLA.md` requires every contributor to sign before a first merge. It is unusually candid about why: "For the project to be able to offer a hosted edition and a commercial license alongside the open source release, one party has to hold the right to relicense the combined work." Contributors keep copyright, but grant Dpro GmbH a "perpetual, worldwide, non-exclusive, royalty-free, irrevocable" license that explicitly "includes the right to relicense Your Contribution, including under proprietary or commercial license terms."

Nothing here is deceptive. It is a standard arrangement used by plenty of commercial open source companies. But "open source" in the video's framing implies community ownership, and the actual arrangement is that one Vienna GmbH holds the commercial rights to everything anyone contributes, including the 30 open `good-first-issue` and translation tasks the README recruits for.

The security notes make a legal claim worth checking. The README says "Recording announcements are on by default. Austria requires both parties to be aware that a call is recorded." The default is real, `RECORDING_ANNOUNCEMENT=true` in `.env.example`. The legal framing is looser than the statute. Austrian StGB § 120 penalizes using a recording device to obtain knowledge of a non-public statement not intended for the recorder, with up to one year imprisonment or 720 daily rates. Practitioner guidance treats Austria as an all-party consent jurisdiction, requiring consent rather than mere awareness. An announcement is a reasonable implementation and it is more than most projects ship, but "aware" and "consented" are not the same word, and US operators fall under their own state all-party consent rules that this default was not designed around.

## Key Takeaways

- The repository is real and verified: `Dpro-at/Tel-Agent`, created 2026-08-16, AGPL-3.0, 516 stars, 115 forks, 210 commits, 30 open issues, releases `v0.1.0` and `v0.1.1` both on 2026-09-05. The video does not misstate what the repo is.
- The video's claim that it "runs entirely on your own hardware" does not hold for the shipped configuration. The default telephony mode points at LiveKit Cloud, and the default STT, LLM, and TTS endpoints are `api.deepgram.com`, `api.openai.com`, and `api.elevenlabs.io`. The project's own blog concedes "the design that gets the first call answered uses hosted speech-to-text, a hosted model and a hosted voice."
- The "Product of the day, 1st" seal visible in every frame is a local SVG. The repository's own `docs/brand/awards/README.md` states it belongs to Flowxtra, a different Dpro product, and that Tel-Agent "has not placed anywhere." I found no Product Hunt listing for Tel-Agent, and I could not independently verify the original Flowxtra award either.
- The "DSGVO & GDPR COMPLIANT" and "AI ACT COMPLIANT" badges are also reused Flowxtra artwork. The repository's own compliance README says these are "properties of a deployment rather than of a repository."
- The phone call feature the video leads with is milestone 11 of 12 on the project's published roadmap, with all six of its acceptance checks unticked. The project is at milestone 3. The transport code exists, the live-line measurement does not.
- "24 channels" is scope, not shipped surface. Nine messaging channels plus phone transport are wired. Seven channel issues were still open when I checked on 2026-09-15.
- Contribution is concentrated: 148 of 210 commits from one account, `9mtm`, who commits as Mohamed Alarade, the managing director named in the CLA. Weekly commit counts peaked at 84 and 86 in consecutive weeks, and the README credits AI coding agents.
- Every contributor signs a CLA granting Dpro GmbH an irrevocable right to relicense their work commercially, which the CLA states is to enable a hosted edition and a paid license.
- The security defaults are the genuinely good part and the video never mentions them: loopback-only binding, no default credentials, API keys encrypted at rest, recording announcement on by default.
- I could not verify the Austrian legal framing in the README beyond secondary sources. StGB § 120 as published on RIS prohibits unauthorized recording of non-public statements with up to one year imprisonment, and practitioner guidance reads Austria as requiring consent from all parties, which is stricter than the README's word "aware."

## Resources

- [Dpro-at/Tel-Agent on GitHub](https://github.com/Dpro-at/Tel-Agent) - The repository the video scrolls, AGPL-3.0, 516 stars when I checked on 2026-09-15.
- [Tel-Agent README](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/README.md) - The exact text the voiceover paraphrases, including the architecture table naming Deepgram, one cloud LLM, and ElevenLabs as the v1 providers.
- [Tel-Agent `.env.example`](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/.env.example) - Shows `TELEPHONY_MODE=livekit` as the default and the three hosted provider base URLs.
- [Tel-Agent award seal documentation](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/docs/brand/awards/README.md) - The file stating the "Product of the day" wreath is Flowxtra's and that Tel-Agent has not placed anywhere.
- [Tel-Agent compliance badge documentation](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/docs/brand/compliance/README.md) - Explains the GDPR and AI Act badges are reused Flowxtra artwork and should not be read as certification.
- [Tel-Agent ROADMAP.md](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/docs/ROADMAP.md) - Twelve milestones, current position at milestone 3, phone at milestone 11 with six unticked checks.
- [Tel-Agent CLA.md](https://raw.githubusercontent.com/Dpro-at/Tel-Agent/main/CLA.md) - The contributor agreement granting Dpro GmbH relicensing rights, and the company's full registration details.
- [The GDPR case for a self-hosted AI phone assistant](https://tel-agent.com/blog/gdpr-self-hosted-ai-phone-assistant) - The project's own post conceding that the current design uses hosted speech, model, and voice.
- [tel-agent.com](https://tel-agent.com) - Project site and download page linked from the README.
- [Dpro GmbH](https://dpro.at/en/about-us/) - The Vienna company behind the project, listing both Flowxtra and the AI phone assistant in its portfolio.
- [livekit/agents](https://github.com/livekit/agents) - The Apache 2.0 voice agent framework Tel-Agent builds on, 14,205 stars, created 2023-10-19.
- [EU AI Act Article 50](https://artificialintelligenceact.eu/article/50/) - The transparency obligation requiring disclosure that a person is interacting with an AI system.
- [Austrian StGB § 120 on RIS](https://www.ris.bka.gv.at/NormDokument.wxe?Abfrage=Bundesnormen&Gesetzesnummer=10002296&Paragraf=120) - The statute behind the README's recording claim, penalizing unauthorized recording of non-public statements.
- [Open Source AI Phone Assistant on DEV](https://dev.to/9mtm/open-source-ai-phone-assistant-call-chat-25oe) - The maintainer's own August 2026 writeup of the project, which makes no award claim.
