Prism

Self-hosted AI that reads the internet so you don't have to.
Running 24/7 on a Mac Studio. 81 sources, 760 signals/week, $0 in LLM API fees.
81
信号源 · sources
4247 /wk
原始条目 · items ingested
760 /wk
LLM 提炼信号 · signals distilled
597
high-value (79%)
142
自主决策 · autonomous decisions /30d
35
天连续运行 · days running
$15.96 $0 / week
同样工作量(约 2.3M tokens/周)在 Claude Sonnet 4.5 上 每年要 ~$832.0。 Prism 全部跑在本地 omlx,LLM 成本 $0,数据不出设备。

Active sources · 来源分布

x 46 youtube 10 xiaoyuzhou 10 youtube_home 2 x_home 1 producthunt 1 model_economics 1 hn_search 1 hackernews 1 github_trending 1 github_releases 1 github_home 1 git_practice 1 follow_builders 1 course 1 claude_sessions 1 arxiv 1

Top signals this week · 本周高价值信号

★★★★★ OpenAI models coming to Amazon Bedrock: Interview with OpenAI and AWS CEOs source ↗
OpenAI 模型将正式集成至 Amazon Bedrock 云服务平台。
Why it matters: 改变云服务 AI 竞争格局,强化 AWS 生态竞争力。
strategic · merged 1 item · 2026-04-29
★★★★★ GitHub RCE Vulnerability: CVE-2026-3854 Breakdown source ↗
GitHub 远程代码执行漏洞 (CVE-2026-3854) 的技术深度解析。
Why it matters: GitHub 核心平台存在 RCE 漏洞,直接威胁软件供应链安全。
actionable · merged 1 item · 2026-04-29
★★★★★ @sama: we have updated our partnership with microsoft. microsoft will remain our prima source ↗
OpenAI 更新与微软合作,维持微软为首选云伙伴,但支持多云部署,合作期至 2032 年。
Why it matters: 标志着 OpenAI 转向多云策略,提升了生态灵活性与市场竞争力。
strategic · merged 1 item · 2026-04-28
★★★★★ @petergyang: My top 5 takeaways from @tibo_maker, a solo AI founder who's making $1M+ a month source ↗
月入百万美元 AI 独立开发者的 5 个实操经验:首日收费、随需求转型、高客单价、低流失率及 SEO 策略。
Why it matters: 为 AI SaaS 创业者提供从定价到增长的具体实操指南。
actionable · merged 1 item · 2026-04-28
★★★★★ China's DeepSeek prices new V4 AI model at 97% below OpenAI's GPT-5.5 source ↗
DeepSeek V4 模型定价仅为 OpenAI GPT-5.5 的 3%,极具价格竞争力。
Why it matters: 极低定价将颠覆 AI 模型市场竞争格局与成本结构。
strategic · merged 1 item · 2026-04-28
★★★★★ China orders Meta to unwind $2B buy of AI startup Manus source ↗
中国要求 Meta 撤回对 AI 初创公司 Manus 的 20 亿美元收购。
Why it matters: 预示中外 AI 领域跨境并购监管风险显著增加。
strategic · merged 1 item · 2026-04-27
★★★★★ Andrej Karpathy: How I use LLMs [video] source ↗
Andrej Karpathy 分享其使用大语言模型(LLM)的具体方法与工作流。
Why it matters: 顶级 AI 专家实战工作流的深度参考。
actionable · merged 1 item · 2026-04-27
★★★★★ Sawe becomes first athlete to run a sub-two-hour marathon in a competitive race source ↗
Sabastian Sawe 在正式比赛中首次实现马拉松破两小时的历史性突破。
Why it matters: 标志着人类生理极限的重大突破与竞技体育新纪元的开启。
strategic · merged 1 item · 2026-04-27
★★★★★ @sama: GPT-5.5 and GPT-5.5 Pro are now available in the API! source ↗
OpenAI 正式通过 API 提供 GPT-5.5 及 Pro 版本。
Why it matters: 模型迭代将直接改变开发者应用能力与行业技术基准。
actionable · merged 1 item · 2026-04-25
★★★★★ DeepSeek v4 source ↗
DeepSeek 发布 V4 模型及其 API 文档,并在 HuggingFace 提供相关资源。
Why it matters: 核心大模型迭代,直接影响开发者工具链与行业竞争格局。
strategic · merged 1 item · 2026-04-25

Recent autonomous decisions · 系统正在自我调整 · full log →

04-28 22:00 x_follow_scan scanned, +0 new, 0 orphan, 0 truncated
04-27 22:00 x_follow_scan scanned, +0 new, 0 orphan, 0 truncated
04-26 22:00 x_follow_scan scanned, +0 new, 0 orphan, 0 truncated
04-25 22:00 x_follow_scan scanned, +0 new, 0 orphan, 0 truncated
04-24 22:00 x_follow_scan scanned, +0 new, 0 orphan, 0 truncated
04-23 22:00 x_follow_scan scanned, +0 new, 0 orphan, 0 truncated
04-22 23:03 adjust_source_weight hn:search: 0.40 -> 0.20, win_rate=0.00
04-22 22:00 x_follow_scan scanned, +0 new, 0 orphan, 0 truncated
04-21 22:19 adjust_source_weight hn:search: 0.60 -> 0.40, win_rate=0.00
04-21 22:00 x_follow_scan scanned, +0 new, 0 orphan, 0 truncated
04-20 22:00 x_follow_blocked bird exited 1: (node:53252) [UNDICI-EHPA] Warning: EnvHttpProxyAgent is experimental, expect them to change at any time. | (Use `node --trace-warnings ...` to show where the warning was created) | [err] Failed to fetch following: HTTP 503:
04-20 12:37 x_follow_blocked bird exited 1: [err] Failed to get current user: fetch failed

How it works · pipeline

sources.yaml
    │    136 × {x, youtube, xiaoyuzhou, reddit, arxiv, github, …}
    ▼
[sync]       fetch raw items
    │
    ▼
[cluster]    dedup + semantic grouping
    │
    ▼
[analyze]    local LLM → summary · why · strength · tags
    │
    ▼
[pairwise]   Bradley-Terry + your 1-bit choices → preference model
    │
    ▼
[decision]   adjust source weights · pick new sources · flag anomalies
    │        (everything logged to decision_log for auditability)
    ▼
signals → /feed · /briefing · /showcase
  
Prism is open source · ⭐ Star on GitHub · about