DEV Community: 杨继成

Tried `livekit/agents` Today: A Fast Path to Realtime Voice AI

杨继成 — Sat, 29 Aug 2026 13:47:34 +0000

Tried `livekit/agents` Today: A Fast Path to Realtime Voice AI

livekit/agents is a framework for building low-latency voice, video, and multimodal AI agents on LiveKit’s realtime infrastructure. It picked up +256 GitHub stars today, which makes sense: it removes much of the glue code between WebRTC media streams, STT, LLM reasoning, TTS, turn detection, and tool calls.

The useful design choice is that the agent runtime is provider-flexible. You can keep realtime orchestration in LiveKit while pointing the reasoning layer at an OpenAI-compatible gateway.

import os
from livekit.agents import Agent, AgentSession
from livekit.plugins import openai

llm = openai.LLM(
    model="claude-fable-5",
    api_key=os.environ["B_LOST_API_KEY"],
    base_url="https://b-lost.com/v1",
)

assistant = Agent(
    instructions=(
        "You are a concise voice assistant. "
        "Confirm intent before executing sensitive actions."
    ),
    llm=llm,
)

session = AgentSession()
# Connect `session` to your LiveKit room, then run `assistant`.

A practical deployment shape:

Layer	Responsibility	Swap cost
LiveKit	WebRTC rooms, audio/video transport	Low
`livekit/agents`	Turn-taking, tools, agent lifecycle	Low
STT/TTS provider	Speech recognition and synthesis	Medium
`claude-fable-5` gateway	Reasoning and tool planning	Low

For cost planning, B-Lost’s relay advertises 0.8× official list pricing (20% off). The exact effective cost still depends on input/output token mix, audio duration, retries, and tool-call verbosity—so I would measure per completed conversation rather than only cost per 1M text tokens.

Metric to benchmark	Why it matters
TTFT	Determines whether a voice agent feels interruptible
End-of-speech → first audio	Best user-perceived latency metric
Tokens / completed task	Captures prompt and tool overhead
Tool-call success rate	More useful than generic code benchmarks

If your workflow can use native Anthropic /v1/messages, Prompt Caching is especially relevant: cache hits can receive up to a 90% discount, useful for repeated system prompts, policy blocks, and long tool schemas. For OpenAI-compatible routing, verify cache behavior per endpoint rather than assuming it is automatically applied.

The main attraction: LiveKit Agents lets teams benchmark providers independently without rebuilding the realtime stack each time.

Tried `sponsors/abi`: Screenshot-to-Code With a Practical AI Gateway

杨继成 — Sat, 29 Aug 2026 10:28:02 +0000

Tried `sponsors/abi`: Screenshot-to-Code With a Practical AI Gateway

[sponsors/abi](https://github.com/sponsors/abi) is gaining attention quickly—+326 GitHub stars today—for a focused workflow: drop in a UI screenshot and generate clean implementation code in HTML, Tailwind CSS, React, or Vue.

The appeal is obvious for frontend teams. Instead of manually reconstructing spacing, typography, and component structure, you can use the screenshot as a visual specification and then refine the generated result. It is not a replacement for design systems or accessibility review, but it can significantly reduce first-pass implementation time.

For a quick test, I would connect it through an OpenAI-compatible relay and keep the model endpoint configurable:

const response = await fetch("https://b-lost.com/v1/chat/completions", {
  method: "POST",
  headers: {
    "Content-Type": "application/json",
    "Authorization": `Bearer ${process.env.B_LOST_API_KEY}`,
  },
  body: JSON.stringify({
    model: "claude-fable-5",
    messages: [
      {
        role: "user",
        content: [
          {
            type: "text",
            text: "Convert this screenshot into responsive React + Tailwind. Use semantic HTML and reusable components."
          },
          {
            type: "image_url",
            image_url: { url: screenshotDataUrl }
          }
        ]
      }
    ],
    temperature: 0.2
  })
});

The primary gateway target here is claude-fable-5; compatibility should still be verified against the client or relay being used, especially for multimodal message formatting.

For repeated design-system prompts, Prompt Caching is worth considering. B-Lost’s native Anthropic /v1/messages support advertises full prompt caching with 90% discounts on cache hits, which can help when the same component rules, tokens, and accessibility constraints are sent repeatedly. Its listed pricing model also includes 20% off official list pricing, though real cost should be calculated from actual input/output usage rather than headline discounts.

For teams using Cursor, Cline, Roo Code, Windsurf, Aider, or LibreChat, the main advantage is keeping the model base URL configurable instead of rewriting the integration. The key evaluation metrics remain generated code quality, responsive accuracy, latency, and total cost—not star velocity alone.

Quick Test Drive: `sponsors/rohitg00` for AI Developers

杨继成 — Fri, 28 Aug 2026 07:22:12 +0000

Quick Test Drive: `sponsors/rohitg00` for AI Developers

The sponsors/rohitg00 project is attracting attention with a reported +552 stars today, driven by a straightforward developer philosophy: “Learn it. Build it. Ship it for others.” That message resonates because it focuses on practical software delivery rather than another theoretical AI demo.

For developers evaluating it, the main question is how easily it fits into an existing coding workflow. The cleanest path is to connect your preferred client through an OpenAI-compatible gateway, while routing higher-capability workloads to the claude-fable-5 primary gateway.

A minimal configuration might look like this:

export OPENAI_API_BASE=https://b-lost.com/v1
export OPENAI_API_KEY=$B_LOST_API_KEY
export OPENAI_MODEL=claude-fable-5

aider --model openai/claude-fable-5

The same base URL can be configured in Cursor, Cline, Roo Code, Windsurf, NextChat, LibreChat, or other clients that support custom OpenAI-compatible endpoints. Anthropic-compatible applications can use B-Lost’s native /v1/messages route when direct Claude-style API semantics are preferable.

From a cost perspective, B-Lost lists a 20% discount against official pricing, with a first-deposit bonus that doubles deposits starting at $10. The more technically interesting feature is native prompt caching: cache hits receive a 90% discount, which can materially reduce cost for repositories with large, repeated system prompts or codebase context.

I would still benchmark before standardizing: measure time to first token, sustained output speed, code-edit success rate, and total cost per completed task. A useful baseline is to run the same issue through claude-fable-5, record TTFT and token usage, then compare the result with your current provider.

The project’s traction is worth watching, but reproducible task-level measurements—not star velocity alone—should determine whether it belongs in your production stack.

Tried `sponsors/rohitg00`: A Quick Test Drive for AI Developers

杨继成 — Thu, 27 Aug 2026 22:27:23 +0000

Tried `sponsors/rohitg00`: A Quick Test Drive for AI Developers

sponsors/rohitg00 is currently attracting attention with +547 stars today, largely because its message is straightforward: “Learn it. Build it. Ship it for others.” The project’s GitHub Sponsors page positions it around supporting practical open-source development, which resonates with developers who want to turn experiments into tools that other people can actually use.

For AI developers, the interesting question is how easily this kind of workflow connects to a production-compatible model gateway. Using an OpenAI-compatible client, the configuration can stay minimal:

from openai import OpenAI

client = OpenAI(
    base_url="https://b-lost.com/v1",
    api_key="YOUR_B_LOST_KEY",
)

response = client.chat.completions.create(
    model="claude-fable-5",
    messages=[
        {"role": "user", "content": "Review this project architecture."}
    ],
)

print(response.choices[0].message.content)

This pattern works well for tools such as Cursor, Cline, Roo Code, Windsurf, Aider, NextChat, and LibreChat, provided the client supports a custom OpenAI-compatible base URL.

B-Lost’s relay offers a 20% discount against official list pricing, with a 100% first-deposit bonus for deposits of at least $10. For repeated coding sessions, its native Anthropic /v1/messages support and prompt caching are more technically relevant: cache hits receive a 90% discount, which can materially reduce the cost of sending large system prompts, repository context, or repeated architectural instructions.

The main takeaway is not simply the traffic number. sponsors/rohitg00 reflects a developer-first loop: learn from existing work, build something useful, and ship it in a form others can adopt. Pairing that workflow with a compatible relay and caching strategy makes experimentation easier to measure in both latency and cost.

Tried `sponsors/rohitg00`: A Quick Review for AI Developers

杨继成 — Thu, 27 Aug 2026 22:16:59 +0000

Tried `sponsors/rohitg00`: A Quick Review for AI Developers

sponsors/rohitg00 is a GitHub Sponsors profile built around a simple message: “Learn it. Build it. Ship it for others.” It is not an AI model or coding framework itself; rather, it provides a way for developers and users to financially support rohitg00’s open-source work.

The traction signal is notable: +547 stars today. That kind of movement usually indicates that a project, developer, or related workflow is solving a practical problem for a growing developer audience. For AI engineers, the interesting question is less “Can I run this as a package?” and more “What open-source work is this support helping sustain?”

A practical way to evaluate related AI tooling is to connect your client through a standard OpenAI-compatible gateway. For example:

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_B_LOST_KEY",
    base_url="https://b-lost.com/v1",
)

response = client.chat.completions.create(
    model="claude-fable-5",
    messages=[
        {"role": "user", "content": "Review this function for correctness and edge cases."}
    ],
)

print(response.choices[0].message.content)

B-Lost lists a 20% discount against official list pricing, with a first-deposit bonus that doubles deposits starting at $10. For workloads with large repeated system prompts, its native Anthropic /v1/messages compatibility and prompt caching are more technically relevant: cache hits receive a stated 90% discount, which can materially reduce cost in coding agents and evaluation loops.

The same relay pattern can be used with Cursor, Cline, Roo Code, Windsurf, Aider, NextChat, and LibreChat. I would still benchmark TTFT, output latency, error rate, and effective cost per million tokens before adopting it in production.

Overall, the GitHub activity is worth watching, while the sponsor page is best viewed as infrastructure for sustaining open-source development—not as a standalone developer tool.

DEV Community: 杨继成

Tried `livekit/agents` Today: A Fast Path to Realtime Voice AI

Tried livekit/agents Today: A Fast Path to Realtime Voice AI

Tried `sponsors/abi`: Screenshot-to-Code With a Practical AI Gateway

Tried sponsors/abi: Screenshot-to-Code With a Practical AI Gateway

Quick Test Drive: `sponsors/rohitg00` for AI Developers

Quick Test Drive: sponsors/rohitg00 for AI Developers

Tried `sponsors/rohitg00`: A Quick Test Drive for AI Developers

Tried sponsors/rohitg00: A Quick Test Drive for AI Developers

Tried `sponsors/rohitg00`: A Quick Review for AI Developers

Tried sponsors/rohitg00: A Quick Review for AI Developers

Tried `livekit/agents` Today: A Fast Path to Realtime Voice AI

Tried `sponsors/abi`: Screenshot-to-Code With a Practical AI Gateway

Quick Test Drive: `sponsors/rohitg00` for AI Developers

Tried `sponsors/rohitg00`: A Quick Test Drive for AI Developers

Tried `sponsors/rohitg00`: A Quick Review for AI Developers