Problem
The streamcoreai/examples repo proves the SDKs work. It does not prove the positioning.
What is there today is one shape repeated per language: connect to a server, talk to an LLM agent, see a transcript. typescript, python, golang, golang-tui, rust, rust-tui, esp32, esp32-desktop-car, bring-your-own-agent. Useful as SDK smoke tests, and every one of them says the same thing about what this project is.
The claim being made is realtime media infrastructure — a WebRTC transport with a pluggable pipeline on top, where the LLM is one thing you can put in the middle rather than the point of the exercise. Nothing in the examples repo demonstrates that. A visitor reasonably concludes this is a voice-chatbot starter, because every example is a voice chatbot.
Proposed change
Six examples, each chosen because it is the smallest thing that makes a different claim. Each is independently useful, so these can land one at a time and this issue can be split.
1. Realtime translator. Caller speaks one language, agent replies in another, both transcripts shown. Demonstrates that the pipeline is a transform, not a chat loop, and it is the single most convincing thing this stack can do in a 30-second video.
2. AI-hosted voice room. Multiple human participants plus the agent as a participant that listens continuously and speaks when addressed. This is the one that most directly contradicts "it's a chatbot", and it will surface real design questions about turn-taking with more than two parties — worth knowing about either way.
3. Browser copilot. The agent sees page context and drives the UI through the DataChannel. Uses the existing tool/plugin path and the vision message assembly in internal/pipeline/vision.go, which no example currently exercises.
4. Embedded device. A finished ESP32-S3 build with a specific parts list and a flashing guide, rather than a firmware source tree. The current ESP32 examples require the reader to already own the hardware and know ESP-IDF.
5. SIP application. streamcoreai/sip-server exists; no example calls it. A phone number that reaches an agent, with the PSTN setup written down. Note the echo-reference caveat from #48 while that is open.
6. Raw audio processing, no LLM at all. The important one. A WHIP endpoint that runs a DSP transform — pitch shift, noise gate, a recorder, an energy meter — with no STT, no LLM, no TTS configured. If the server cannot be configured to do this cleanly today, that is a finding worth having, and it is exactly the claim "media infrastructure, not framework" rests on.
Conventions for each, so they stay maintainable rather than rotting into a second codebase: a README with a one-command run, a config.toml fragment showing only what differs from the default, a screenshot or short clip, and no dependency on credentials beyond what the quick start already asks for. Cross-link from the main README's example table and from docs/capabilities.md.
Per repo convention, README updates ship with the example, including the zh-CN mirrors.
Acceptance criteria
Pointers
Problem
The
streamcoreai/examplesrepo proves the SDKs work. It does not prove the positioning.What is there today is one shape repeated per language: connect to a server, talk to an LLM agent, see a transcript.
typescript,python,golang,golang-tui,rust,rust-tui,esp32,esp32-desktop-car,bring-your-own-agent. Useful as SDK smoke tests, and every one of them says the same thing about what this project is.The claim being made is realtime media infrastructure — a WebRTC transport with a pluggable pipeline on top, where the LLM is one thing you can put in the middle rather than the point of the exercise. Nothing in the examples repo demonstrates that. A visitor reasonably concludes this is a voice-chatbot starter, because every example is a voice chatbot.
Proposed change
Six examples, each chosen because it is the smallest thing that makes a different claim. Each is independently useful, so these can land one at a time and this issue can be split.
1. Realtime translator. Caller speaks one language, agent replies in another, both transcripts shown. Demonstrates that the pipeline is a transform, not a chat loop, and it is the single most convincing thing this stack can do in a 30-second video.
2. AI-hosted voice room. Multiple human participants plus the agent as a participant that listens continuously and speaks when addressed. This is the one that most directly contradicts "it's a chatbot", and it will surface real design questions about turn-taking with more than two parties — worth knowing about either way.
3. Browser copilot. The agent sees page context and drives the UI through the DataChannel. Uses the existing tool/plugin path and the vision message assembly in
internal/pipeline/vision.go, which no example currently exercises.4. Embedded device. A finished ESP32-S3 build with a specific parts list and a flashing guide, rather than a firmware source tree. The current ESP32 examples require the reader to already own the hardware and know ESP-IDF.
5. SIP application.
streamcoreai/sip-serverexists; no example calls it. A phone number that reaches an agent, with the PSTN setup written down. Note the echo-reference caveat from #48 while that is open.6. Raw audio processing, no LLM at all. The important one. A WHIP endpoint that runs a DSP transform — pitch shift, noise gate, a recorder, an energy meter — with no STT, no LLM, no TTS configured. If the server cannot be configured to do this cleanly today, that is a finding worth having, and it is exactly the claim "media infrastructure, not framework" rests on.
Conventions for each, so they stay maintainable rather than rotting into a second codebase: a README with a one-command run, a
config.tomlfragment showing only what differs from the default, a screenshot or short clip, and no dependency on credentials beyond what the quick start already asks for. Cross-link from the main README's example table and fromdocs/capabilities.md.Per repo convention, README updates ship with the example, including the
zh-CNmirrors.Acceptance criteria
docs/capabilities.mdlink to each, withzh-CNmirrors updated.Pointers
streamcoreai/examples— where these land; follow the existing structure and itsAGENTS.mdconventionsinternal/pipeline/vision.go— vision message assembly, unused by any exampleinternal/plugin/— tool path for the browser copilotstreamcoreai/sip-server— the SIP example's server sidedocs/capabilities.md— the table these examples are meant to make honest