Skip to main content
Reading as

Signal

What actually matters in AI.

Signal pulls the makers’ own announcements, ranks each by Signal Strength, decodes what builders actually do with them, and shows you the real tools to make your own.

Decoded · what builders actually do

The actual playbooks, pulled from the people building.

Tracked makersIndyDevDan84Cole Medin82Systems Made Better80AI Jason80Matthew Berman78David Ondrej76Matt Wolfe74Better Creating72Wes Roth72Riley Brown70
David Ondrej · Decoded

Pi Agent Setup & Configuration Mastery

Install and configure Pi Agent (pi.dev) as a minimal, highly customizable local AI coding agent — connect it to OpenRouter for multi-model support, wire up web search, and tune context files so the agent adapts to your workflow rather than the reverse.

  1. 1. Install Pi globally: copy the one-liner curl command from pi.dev and run it in any terminal (~20 seconds).
  2. 2. Launch Pi by typing `pi` in the terminal; if not authenticated, type `/lo` to open the provider selection menu.
  3. 3. Choose a provider (e.g., OpenRouter): log in to openrouter.ai, add $5–$10 in credits, generate a new API key with a spending limit, paste it into Pi when prompted.
  4. 4. Verify the connection by sending a test message (e.g., 'say hi') and confirming a response.
  5. 5. Select your model with `/model` (e.g., Claude Opus 4.8 Fast); adjust thinking effort at any time with Shift+Tab cycling through off → minimal → low → medium → high → extra high.
  6. 6. Understand Pi's four built-in tools — read, write, edit, bash — and internalize that bash alone gives the agent full terminal control (file management, package installs, network analysis, etc.).
  7. 7. Tune global behavior by asking Pi to locate its global folder and edit the `append-system.md` file; add preferences such as language rules, date format, response conciseness, or 'always relay bash output in text'.
  8. 8. For project-specific context, navigate to the project folder and ask Pi to create an `agents.md` file with relevant context; Pi loads it automatically in that directory.
  9. 9. Add web search capability: run `pip install pi-web-access` (or ask Pi to check and install it) — the package receives 90k+ downloads/month and is the standard extension.
  10. 10. Use Pi itself to manage its own configuration and extensions — ask it to install packages, update context files, or scaffold new agents rather than doing everything manually.

Gotcha Do not touch `system.md` directly — it overrides the entire system prompt and can break default behaviors set by the maintainer. Use `append-system.md` instead to extend behavior safely. Also keep API keys private and set a spending cap on OpenRouter to avoid unexpected charges.

Watch the full video ↗Download build kit ↓✦ Enhanced kit · sign in

A CLAUDE.md build brief + README. Unzip it into an empty repo, open the folder with your coding agent, and tell it to read CLAUDE.md and plan before building. A derived starting point — verify against David Ondrej’s video.

The lead story

Model release· Jun 15 · Signal 68

Anthropic releases Claude Opus 4.8

Anthropic updated its most capable Claude model, Opus 4.8, saying it performs better at coding, multi-step tasks, and professional work. The company also highlights that it handles long tasks more reliably without losing track.

↗ anthropic
Maker claims
Stronger performance across coding, agentic tasks, and professional work; improved consistency for long-running work.
Builders found
No builder feedback or independent evaluations are present in the available sources at this time.
Benchmark
No benchmark data is cited in the source. The announcement is a product page only; no third-party evals or comparison numbers are provided.

This week · ranked by signal

Model release· Jun 11 · Signal 68
☆ Save

Anthropic Announces Claude Fable 5 and Mythos 5

Anthropic has announced two new AI models, Claude Fable 5 and Claude Mythos 5, described as their most capable yet for difficult research and coding tasks. No independent reviews or test results are available yet.

Maker claimsAnthropic describes Fable 5 and Mythos 5 as their next generation of intelligence, built for the hardest knowledge work and coding problems.
Builders foundNo builder feedback, independent evaluations, or real-world deployment reports are present in the available sources.
BenchmarkNo benchmark data is available in the provided sources. The announcement does not cite specific evaluation results.
↗ anthropic

The part no one else has

What do you want to make?

Pick an outcome — get the exact stack: the tool for each step, why it’s the pick, what it costs, and how they hand off. Every tool is clickable.

Make a cinematic AI video clip

The "wait, is that real?" clips — from an idea to a finished shot.

◷Vertical 9:16 for TikTok/Reels/Shorts; 16:9 for YouTube.
  1. 1
    Write the shotFree–$20/mo

    Tighten the concept + a vivid visual prompt.

    Or: Jasper

    Still needs your taste.

  2. 2
    Make the starting image$10–30/mo

    A strong still gives the video model something faithful to animate.

    Or: Nano Banana (Gemini Image)

    Nailing the exact look takes iterations.

  3. 3
    Animate it$10–40/mo

    Sora 2 = most realistic; Veo 3 adds sound; Kling = cheaper + longer; Runway = best fine control.

    Or: Hailuo · Dream Machine · Pika

    Short clips; shot-to-shot consistency is hard.

  4. 4
    Cut + captionFree

    Assemble shots, add captions/music, export to the right ratio.

    Or: Descript

    Advanced effects paywalled.

Combined cost · ~$20–60/mo, mostly the video model.

Heads up · Free tiers usually watermark; check each model's commercial-use terms; label as AI where the platform requires.

Model Index

Which model, and what it costs.

Alibaba · active

☆ Save

Qwen 3 Max

Input price
$1.20/M
what you send
Output price
$6.00/M
what it writes back
Context
256K
reads at once

Priced per 1M tokens (~750,000 words) — a typical back-and-forth runs about under 1¢, so even heavy daily use is only a few dollars a month.

Best for multilingual · coding

Approximate pricing — verify.

Pricing history — track how this model’s price + status move over time. Sign in — free →

Prices are per 1M tokens and change often — verify before relying on them.

Export CSV ↓

Want one of these built into your own business?

Start a project →