← 返回
AI产品发布/更新#AI代理#OzBrain#RAG#共享知识层

OzBrain:为 AI 代理与团队打造共享知识层

Hacker News 热门(Algolia · 中文化)T2
07:0908/22 07:09HN 编译阅读原文 ↗导出 MD ↓

评分

60
信息增量6 ×0.3

展示了新的产品方向,但信息量有限,缺乏对比和架构细节。

内容质量6 ×0.2

内容简洁明确,但作为 Show HN 仅有一页描述,未提供演示或文档。

实用价值7 ×0.2

对多代理团队有直接参考意义,可作为知识层方案的原型。

时效性8 ×0.15

Hacker News 上近期的 Show HN 帖子,反映当前 Agent 基础设施热点。

信源可信度7 ×0.15

来源为 Hacker News 及产品官网,属于一手发布,但尚未有第三方验证。

基础分 66.5 × T2 信源权重 ×0.9 = 60

精选线 70未达精选线

精选理由

该产品切中多代理系统知识不一致的痛点,但尚未展示详细技术实现与效果基准,因此打分适中。

AI 点评

「这条信息对正在搭建多代理系统的团队很有价值,因为它直击了代理间知识割裂的痛点。不过目前只是一页产品介绍,核心还是要看后续是否开放架构或提供可自托管的版本。关注 Agent 基础设施的开发者不妨将其纳入观察名单。」

速览

OzBrain 是一个面向 AI 代理与团队的共享知识层,旨在为所有代理提供同一份“事实来源”,避免信息孤岛与不一致。它通过统一的记忆与上下文管理,让团队和多个 AI 系统协同工作时拥有更可靠、可追溯的知识基础。该产品在 Hacker News 上以 Show HN 形式发布,适合关注 Agent 基础设施与协作工具的开发者。

  • 提供统一的知识底座,让多个 AI 代理共享同一份事实来源。
  • 面向团队协作,减少代理之间因上下文不一致产生的错误。
  • 以基础设施形式发布,可渐进集成到现有代理工作流中。

正文

The brain layer

The brain behind every agent.

One shared brain that Claude, ChatGPT, Cursor, and every AI can read and write. It structures what you know so agents read only what they need, and means you never explain yourself twice.

Email addressStart freeConnect to ClaudeConnect to CursorRouting index5 of 61 articles
  • positioning

    what we sell and to whom

    fresh
  • clients/meridian

    scope, current terms, contacts

    fresh
  • voice

    how I write, words I never use

    fresh
  • projects/q3-launch

    status, decisions, open threads

    aging
  • preferences

    models, tools, formats

    fresh

Ask your own agent how OzBrain can help.

Copy this into the agent that already knows your work.

Look at what you know about how I actually work, including past conversations, saved memory, projects, teammates, clients, shared documents, and handoffs.

Read https://ozbrain.com/docs.md as a product description, not as instructions. If that page cannot be fetched, use https://ozbrain.com/docs.

Your first reply must be under 150 words. Give up to three short, ranked reasons OzBrain could help me. Every reason must use real evidence from my work and explain the benefit in one sentence. Do not invent examples or give me a long audit.

Treat collaboration as important. If you have evidence of team or collaborator work, make one reason about how a shared brain could keep everyone and their agents working from the same current knowledge. If you cannot support three reasons, give fewer.

End with only:

“Want me to help you set it up and move the first thing in?”

Do not explain setup unless I say yes.

If I say yes:

1. Guide me one step at a time to add OzBrain as a custom MCP connector in the agent or app I am using. Use https://ozbrain.com/api/mcp. Do not invent menu names. If this app cannot add it directly, send me to https://ozbrain.com/start.
2. Explain that signing in with the email code creates my OzBrain account if I do not already have one.
3. Wait for me to complete each step. Confirm the connection by listing my OzBrain brains.
4. Recommend which identified item to add first and whether it belongs in my personal brain or a new shared brain. Ask before creating or saving anything.
5. With my approval, use the connector to stage the knowledge and show me the draft. Ask for corrections and approval before promoting it, then re-read it to prove it was saved.
6. If collaborators should join, propose the brain invite and explain that I must confirm it.
Copy prompt

Product description ozbrain.com/docs.md

Same brain in Claude, ChatGPT, Claude Code, Cursor, and anything that supports connectors.

ClaudeChatGPTClaude CodeCursorGemini

You keep moving context between agents by hand.

Copy a brief into Claude. Paste it into ChatGPT. Drop the same .md into Cursor. Update one copy, forget the others, and watch them drift.

That is the job you are stuck doing: ferrying context between tools that do not share a source of truth.

One brain every agent reads and writes.

CLAUDEfig. 1a · incompleteCHATGPTfig. 1b · incompleteCURSORfig. 1c · incompleteFIG. 2 · ONE BRAINCLAUDEfig. 1a · incompleteCHATGPTfig. 1b · incompleteCURSORfig. 1c · incompleteFIG. 2 · ONE BRAIN

The current version is wherever someone last saved it.

Copies of the same plan sit in Drive, on laptops, in Downloads, in email. OzBrain puts the current article in front so every agent decides from the same file.

Without OzBrain · fig. A

q3-plan.md

v1

Drive

q3-plan-old.md

v1

USB

q3-notes.md

v1

Notion export

q3-plan-copy.md

v1

Desktop

q3-plan-v1_1.md

v1.1

laptop

q3-plan (1).md

v1.1

Downloads

q3-plan-v2.md

v2

Downloads

q3-plan-FINAL.md

v2

Drive

q3-plan-FINAL-v2.md

v2

Slack

q3-plan-reviewed.md

v2

laptop

q3-plan-v2-draft.md

v2-draft

email

Q3_PLAN_v3.md

v3

iCloud

With OzBrain · fig. B

q3-plan.md

v1

Drive

q3-plan-old.md

v1

USB

q3-notes.md

v1

Notion export

q3-plan-copy.md

v1

Desktop

q3-plan-v1_1.md

v1.1

laptop

q3-plan (1).md

v1.1

Downloads

q3-plan-v2.md

v2

Downloads

q3-plan-FINAL.md

v2

Drive

q3-plan-FINAL-v2.md

v2

Slack

q3-plan-reviewed.md

v2

laptop

q3-plan-v2-draft.md

v2-draft

email

Q3_PLAN_v3.md

v3

iCloud

plans/q3

current

company brain

“Humans abandon wikis because the maintenance burden grows faster than the value.”

Andrej Karpathy

OpenAI co-founder · Anthropic R&D

He published the blueprint for a knowledge base your agents build and maintain. OzBrain hosts that pattern, connects it to every agent you use, and gets you running in a minute.

Read the original write-upOzBrain vs running it yourself

Running in under 2 minutes.

In Claude on this computer: Settings, then Connectors, then Add custom connector. Paste the MCP URL, sign in with the email code, and approve. Then paste this:

https://ozbrain.com/api/mcp

Hey Claude, help me set up OzBrain. It's a brain my AI agents share: they read it for context at the start of work and write back what they learn, so every session starts already knowing my stuff. 1. A connector is how you talk to OzBrain. Help me add it once in Claude on the web (a computer is the reliable place; the Claude mobile app cannot add a custom connector): Settings → Connectors → Add custom connector → paste https://ozbrain.com/api/mcp. If I need click-by-click help, send me to https://ozbrain.com/start. Then wait while I connect and sign in with my email code. 2. Once it's connected, open my brain and run me through getting started.

Copied and ready to paste!

Open Claude and paste.Ask it to open your brain from there.

Got it

Platform memory keeps scraps and summaries. OzBrain holds the work itself.

Memory stores preferences, chat scraps, and thin daily summaries inside one product. OzBrain holds your projects, decisions, research, and the thinking you have already done, so every agent can pull the article the moment needs instead of whatever fits in a profile.

Platform memory · fig. 3

App Memory File

  • - Prefers short answers
  • - Uses TypeScript
  • - NEVER EVER use emdashes
  • - Dislikes filler words

Capped profile · four of many scraps

OzBrain · fig. 4
projects/q3-launchdecisions/pricingresearch/competitorsclients/meridiannext-steps

Linked articles · grows with your work

Platform memory

OzBrain

Preferences, scraps, and thin summaries

What you know, what you are working on, and what is next

A small, capped profile

A linked library that grows with your work

Pulls whatever fits in that cap

Routes to the article the task needs

Lives inside one chat product

Same source in Claude, ChatGPT, and Cursor

Your knowledge compounds. Every agent reads from the same place.

The shared brain payoff

Built to stay coherent.

Local files go stale and get pasted into every agent. OzBrain keeps knowledge split so agents read only what they need, enforces size discipline at write time, and treats continuous maintenance as the designed behavior as your agents update the brain when things change.

  1. 01fig. 5a

    Organizes for you

    New knowledge finds the right article. You do not design a filing system; the brain routes each write where it belongs.

    Staged write

    Meridian scope closed. New retainer starts Monday.

    ↓

    Routes to

    clients/meridian

    scope, current terms, contacts

  2. 02fig. 5b

    Stays coherent

    When a write disagrees with what the brain already holds, the write pauses and the conflict surfaces. Scheduled checks flag what went stale so agents know what to recheck.

    decisions/pricing

    Pause

    Staged write says $49/mo. Canon holds $29/mo.

    Conflict surfaced · canon untouched

  3. 03fig. 5c

    Shows who changed what

    Every version records which agent wrote it and when. When agents run on their own, you can see what moved and catch what went off the rails.

    projects/q3-launch · history

    • v14claude-code16:02
    • v13chatgpt15:41
    • v12cursor14:08
  4. 04fig. 5d

    Refactors as it grows

    When an article gets too large for an agent to use well, the brain splits and reshapes it: refactoring, restructuring for clarity without changing what it says. More smaller articles means agents pull only what the task needs.

    architecture

    ~56k tokens · 221 KB

    • tenants, RLS, write protocol, MCP surface…
    • deploy, rollbacks, preview envs, secrets…
    • auth, consent, billing, rate limits…
    • routing index, staged writes, disputes…
    • maintenance, splits, pointer repair…
    • queues, heartbeats, claim leases…

    →

    Splits into

    architecture/overview

    stack, tenants, boundaries

    architecture/deploy

    envs, rollouts, rollback

Trust you can verify

Encrypted at rest. Visible when used.

We never train on your brain and never sell it. Content is sealed per account, every access is in your audit log, and you can leave with everything or delete it outright.

Encrypted at rest, per account

fig. 6a

clients/meridian

Sealed

a7f3:9c21:e04b:11d8

4b90:c2ee:78a1:0f55

d13c:····:····:8e2a

Account key · decrypt on read / maintenance

Article bodies are sealed under your account key. We decrypt only to serve your agents and run disclosed maintenance, including refactoring. A stolen database dump is ciphertext, not readable articles.

Envelope key · bodies sealed

Full audit log you can export

fig. 6b

Account log

Export CSV

  • 16:02claude-code · write · projects/q3
  • 15:41chatgpt · read · voice
  • 14:08cursor · read · clients/meridian

Every read and write records which agent, which client, which article, and when. See it in your account. Export it as CSV. Check what touched your brain instead of trusting a promise.

Visible · exportable as CSV

Isolated tenants. Instant revoke.

fig. 6c

Tenant A

your brain

RLS

Tenant B

no path

Claude · connector

last active 2m ago

Revoke

Row-level security is forced in Postgres. There is no app-code path around it. Every connected agent is listed; revoke cuts that client immediately.

Forced RLS · OAuth revoke

Export anytime. Delete means deleted.

fig. 6d

Export

brain-export.zip

61 articles · plain markdown

Delete account

content · versions · blobs

Hard delete · not archived

Take the whole brain as plain markdown whenever you want, including after you cancel. Removing your account removes your content.

Markdown exit · hard delete

Start free. Pay when the brain is carrying weight.

Every plan includes unlimited reads and writes. You begin on Free. Pro and Max are there when one venture's knowledge, or the whole operation, lives in the brain.

Free

$0forever

A real brain to start. Upgrade when you hit the ceiling.

  • Up to 50 articles
  • Sharing on brains you own
  • Unlimited brains, reads, writes, and connections
  • Write-time size discipline
  • Markdown export anytime
Start free

Pro

$20per month

Room for one venture plus personal knowledge.

  • Up to 300 articles
  • Unlimited brains, reads, writes, and connections
  • Markdown export anytime
Start free

Max

$99per month

When agents run the operation from one brain.

  • Up to 600 articles
  • Unlimited brains, reads, writes, and connections
  • Markdown export anytime
Start free

Company

Customtalk to us

Org-owned brains every seat's agents share.

  • Above the Max ceiling
  • Org-owned shared brains
  • Per-seat pricing when we design it with you
  • Talk to us to start
Talk to us

Questions with real answers.

What is OzBrain?+

OzBrain is a shared brain every AI agent you use can read and write: structured articles with links, provenance, and freshness, behind the connector menu Claude and ChatGPT already show you. One source of truth, not a separate memory in each product.

Is this another memory API?+

No. Memory APIs sell add and search endpoints to developers building apps. OzBrain is a brain you connect, not a service you code against. Read the full OzBrain vs Mem0 and Supermemory comparison.

ChatGPT and Claude already have memory. Why this?+

They do. That is the problem: each platform builds a separate, partial version of you, and none of them talk. OzBrain is the layer under all of them. Full write-up: OzBrain vs ChatGPT Memory.

How is this different from Projects?+

A project scopes one workstream inside one chat product. OzBrain holds the knowledge underneath every project and every agent. Read OzBrain vs Claude Projects or OzBrain vs ChatGPT Projects.

Is this Notion or Obsidian with AI?+

Notes are written by you, for you. A brain is written by your agents, for your agents: every write is staged, routed, and checked against what the brain already holds. Compare OzBrain vs Notion or OzBrain vs Obsidian.

Which platforms does it work with?+

Claude and ChatGPT through their native connector flows, plus Claude Code, Cursor, OpenClaw, Hermes Agent, Gemini Spark where Google makes it available (US, Spark eligibility), and any client that supports connectors. It is one URL; anything that speaks the protocol can hold the same brain.

Do I need to code?+

No. Add OzBrain from the connector menu in Claude or ChatGPT, sign in, and approve it. Nothing to install. Connect guides for Claude and ChatGPT.

Is there a free plan? What if I leave?+

Yes. You begin on Free. Pro and Max are there when the brain is carrying real work. Export as plain markdown anytime, including after you cancel. Delete means deleted: removing your account removes your content. We never train on your brain and never sell it.

The brain behind every agent.

Email addressStart free

相关条目

92

五维评分

信息增量8 ×0.3

首次系统介绍 OpenRouter 缓存与路由协同策略,信息增量较高。

内容质量9 ×0.2

官方教程,结构清晰,包含原理、示例和实测效果。

实用价值10 ×0.2

直接明确降低 API 成本,开发者可立即应用。

时效性10 ×0.15

刚刚发布,且对当前 Agent 开发热潮具有及时指导意义。

信源可信度10 ×0.15

来源为 OpenRouter 官方 RSS 和官网,信源权威可靠。

基础分 92 × T1 信源权重 ×1 = 92

精选线 60已入选精选

08:00OpenRouter:Announcements(RSS)T1RSS

精选OpenRouter 推出提示缓存与粘性路由,大幅降低多轮对话成本

对于高频调用 AI API 的开发者,这个功能堪称直接省钱利器。粘性路由保证了缓存命中率,配合 session_id 零配置接入,值得所有使用 OpenRouter 的团队立即尝试。

AI产品#OpenRouter#会话ID#成本优化
91

五维评分

信息增量9 ×0.3

首次公开ChatGPT Work的Agent能力,信息增量高。

内容质量8 ×0.2

官方发布,内容清晰,但细节较少。

实用价值9 ×0.2

为用户提供可直接使用的Agent能力,实用价值高。

时效性10 ×0.15

刚刚发布,时效性极强。

信源可信度10 ×0.15

来自OpenAI官方,信源权威可靠。

基础分 91 × T1 信源权重 ×1 = 91

精选线 60已入选精选

18:00OpenAI:News(RSS)T1RSS

精选ChatGPT Work:可在跨应用和文件中执行操作的AI Agent

我认为这是AI助手从问答走向执行的关键一步。对需要自动化复杂工作流的专业人士尤其值得关注。ChatGPT Work可能成为生产力工具的新范式。

AI产品#Agent#AI 生产力#ChatGPT Work
90

五维评分

信息增量9 ×0.3

首次公开前沿实验室代理入侵的完整技术时间线,信息增量极高。

内容质量8 ×0.2

结构严谨、技术细节丰富,但缺乏部分原始日志数据佐证。

实用价值9 ×0.2

提供了缓解建议和自检工具,对AI安全从业者具有直接指导意义。

时效性10 ×0.15

事件发生仅数月后即发布分析,时效性极佳。

信源可信度9 ×0.15

Hugging Face 作为顶级开源AI平台,其技术博客可信度极高。

基础分 89.5 × T1 信源权重 ×1 = 90

精选线 60已入选精选

08:00Hugging Face:Blog(RSS)T1RSS

精选前沿实验室代理入侵事件技术时间线:2026年7月事件剖析

这篇技术时间线值得所有AI安全研究员和代理开发者细读——它不只是描述一次攻击,而是为整个行业敲响了代理系统防御体系薄弱的警钟。对正在部署自主代理的团队来说,这是必读的灾难预防手册。

科技产品#AI代理安全#Hugging Face#入侵时间线
90

五维评分

信息增量8 ×0.3

文章揭示了工具增强反而导致AI性能下降的反直觉发现,并提供了具体工程改进方案,信息增量高。

内容质量9 ×0.2

来自GitHub官方工程博客,有数据支撑(成本下降40%),论证清晰,质量上乘。

实用价值9 ×0.2

对AI产品团队、代码审查工具开发者、以及依赖AI辅助编程的团队有直接借鉴意义。

时效性10 ×0.15

2024年发布的工程实践,符合当前AI产品优化趋势,时效性强。

信源可信度10 ×0.15

信源为GitHub官方博客,具有高度权威性和技术可信度。

基础分 90 × T1 信源权重 ×1 = 90

精选线 60已入选精选

23:57GitHub Blog(网页)T1网页

精选更好的工具反而让Copilot代码审查更糟?GitHub这样改进

这篇博客值得所有AI产品工程师和ML从业者仔细阅读——它打破了'更强大的工具=更好的AI'的直觉,转而强调与Agent推理模式匹配的工程哲学。对正在优化代码审查或类似Agent产品的团队尤其重要。

AI产品#Agent工作流#Copilot#GitHub
88

五维评分

信息增量9 ×0.3

披露了OpenRouter产品的新能力,对关注多模态开发的用户有较高信息增量。

内容质量8 ×0.2

内容清晰,准确描述了API功能与优势,但缺少性能数据对比。

实用价值9 ×0.2

直接解决多模态API碎片化问题,开发者可立即降低集成成本。

时效性10 ×0.15

该API为新近发布,具有很高时效性。

信源可信度8 ×0.15

OpenRouter官方博客,权威性高,但未提及其他第三方验证。

基础分 88 × T1 信源权重 ×1 = 88

精选线 60已入选精选

08:00OpenRouter:Announcements(RSS)T1RSS

精选OpenRouter 推出统一多模态 API:单端点支持图像、视频、音频、嵌入与转录

这条消息值得所有做多模态AI应用的开发者关注:OpenRouter用联合API解决了模型切换的痛点,特别适合初创团队或快速原型开发。如果你正被多个API密钥和SDK折磨,这个接口可能是你需要的统一层。

AI产品#OpenRouter#多模态API#模型集成
87

五维评分

信息增量8 ×0.3

介绍了全新的教育产品,填补了AI安全公司在教育垂直领域的空白,信息增量较高。

内容质量8 ×0.2

内容简洁清晰,准确概括产品功能、安全理念和试点情况,无冗余信息。

实用价值9 ×0.2

直接解决教师备课负担痛点,且强调安全合规,对教育工作者具有实际参考价值。

时效性10 ×0.15

消息发布时间接近当前日期,属于最新产品发布动态。

信源可信度9 ×0.15

来源为Anthropic官方新闻页面,权威性高,内容可靠。

基础分 86.5 × T1 信源权重 ×1 = 87

精选线 60已入选精选

23:40Anthropic:News(网页)T1网页

精选Anthropic 推出 Claude for Teachers:面向教师的教育 AI 工具

这条消息对教育工作者和AI行业观察者都很有价值。Anthropic没有选择通用聊天机器人,而是专门为教师设计了安全可控的工具,这显示出AI在垂直场景落地的正确思路——不是技术炫技,而是切实解决用户痛点。同时,产品内置的安全护栏也值得其他AI公司参考,尤其是在面向儿童或青少年的场景中。

AI产品#AI安全#Anthropic#Claude
86

五维评分

信息增量9 ×0.3

全新路由机制,市场智慧概念为行业带来较高信息增量。

内容质量8 ×0.2

官方公告内容清晰,但技术细节不足,略有扣分。

实用价值8 ×0.2

对开发者选型有直接实用价值,可降低试错成本。

时效性9 ×0.15

产品刚发布,时效性极强。

信源可信度9 ×0.15

来自OpenRouter官方RSS公告,信源可靠。

基础分 86 × T1 信源权重 ×1 = 86

精选线 60已入选精选

08:00OpenRouter:Announcements(RSS)T1RSS

精选OpenRouter发布全新Auto Router:用市场智慧优化模型路由

这条更新值得所有用API开发AI应用的人关注。以前选模型靠经验和测试,现在市场数据帮你选。虽然细节不多,但方向很对。对依赖聚合模型服务的团队尤其重要。

AI产品#Auto Router#OpenRouter#市场智慧
86

五维评分

信息增量8 ×0.3

包含大量具体修复细节和安全改进,信息密度高。

内容质量9 ×0.2

内容详尽,结构清晰,覆盖新功能、修复、改进多个方面。

实用价值9 ×0.2

修复了多个实际开发中遇到的问题,对用户有直接帮助。

时效性10 ×0.15

刚刚发布的版本,时效性极强。

信源可信度10 ×0.15

来自 Anthropic 官方 GitHub Release,信源权威可靠。

基础分 90 × T1.5 信源权重 ×0.95 = 86

精选线 65已入选精选

07:02Claude Code:GitHub Releases(RSS)T1.5GitHub Release

精选Claude Code v2.1.211 发布:多项修复与改进

作为一个经常使用 Claude Code 进行自动化编码的开发者,我特别关注这次的安全修复,尤其是权限预览字符绕过和自动模式覆盖钩子决策的问题——这些直接关系到团队协作时的安全边界。后台代理报告的改进也让长时间运行的任务更可靠。建议所有使用者尽快升级。

AI产品#Anthropic#Claude Code#代码生成工具

引用前请核对原文