Scarlett

• 1. overview (619 chars)

  Scarlett is an AI Co-Worker embedded in Slack and iMessage. It is developed by the tryscarlett.ai team and is driven by the Claude series model at the bottom. The biggest difference between it and common chatbots is that it "does real work": it doesn't just generate text, but calls more than 3,000 application integrations on its own cloud computer to execute multi-step workflows end-to-end - pulling data, writing documents, sending emails, building dashboards, and deploying small applications.   The product positioning is clear and targeted at remote teams, sales and customer service teams, and small and medium-sized enterprises that heavily use Slack. It can use natural language to drive the automation of daily operations.

• 2. Product background (804 chars)

  Scarlett’s official website is scarlett.ai, and its external marketing landing page is tryscarlett.ai. Judging from the page information, the team defines itself as an "AI colleague" rather than an "AI assistant", emphasizing that Scarlett has an independent cloud working environment (cloud computer) like a real colleague, and can maintain context across applications and continue to advance tasks, rather than being limited to answering questions in a single chat thread.   As of the current public information, Scarlett has not disclosed details such as financing rounds, founder background or company registration location. It is an early-stage SaaS entrepreneurial product. It is included in multiple AI tool navigation sites (ExploreAI, MossAI, AIToolly, Dir2AI). The public page claims that "access can be completed in three minutes, can be canceled at any time, and a free credit of US$50 is given." It is a typical self-service PLG route.

• 3. Product functions and usage experience (1957 chars)

  Scarlett’s core capability can be summed up in one sentence: translating natural language instructions into real operations across tools. Users invite Scarlett into a Slack channel or iMessage group and explain the task in plain English, such as "Pull out last week's sales report and send it to #team-updates." It will connect to the corresponding tool, run the workflow, and post the results back into the chat.   Its scale of integration is one of its selling points - officials say it has been connected to more than 3,000 tools, covering CRM (Salesforce, HubSpot), project management (Linear, Notion), calendar, cloud disk, mailbox, payment (Stripe), advertising (Google Ads, Meta Ads), etc. Users can call it without additional configuration after connecting.   What is more important is "execution" rather than "answer". Scarlett runs tasks on her own cloud computer and can draft presentations, produce and send PDFs, write code, build data dashboards, build and deploy small applications, research competing products, and generate detailed reports. For high-risk actions (such as deploying code, sending outgoing emails), it will confirm with the user before executing it, and will self-review the work results.   The product has built-in a number of scenario-based workflows that can be activated with one click, which can intuitively illustrate the boundaries of its capabilities: daily team briefing (summarizing project status and blocking items at 9 a.m. every working day), pre-meeting briefing (grabbing participants, company background and recent email threads), customer feedback topic extraction, customer service work order triage (scanning work orders and emails, classifying issues and summarizing recurring issues), weekly operating reports, drive disk contract/invoice inspection, project release notes (merging GitHub PR and Linear issues Convert to external release instructions and internal changelog), inbox follow-up cleaning (find important Gmail conversations that have not been responded to and draft follow-ups).   At the model level, Scarlett uses Claude as the backbone and selects the most appropriate model according to the task - Opus for chatting, Fable for planning, and Claude 5.6 for coding, so as to take into account the depth of reasoning and execution speed.

• 4. Pricing and business model (712 chars)

  Scarlett has two tiers of pricing. The Team version is $50 per month and includes 25,000 credits, native Slack threads and @mention triggering, all 3,000+ integrations, scheduled and recurring tasks, higher concurrency limits, and advanced automation and drafting capabilities. The Enterprise version is a customized quote that adds SSO and role-based access control, invoices and customized settlement terms, security review and SLA, and exclusive onboarding based on Team.   New users can get $50 in free credits. Officials claim that access takes less than three minutes and can be canceled at any time. It should be noted that some third-party navigation sites (such as ExploreAI) have a starting price of "$29 per month", which is different from the $50 on the official landing page. The actual price should be based on the official website.

• 5. User reviews (1009 chars)

  Judging from the aggregated scores of various tool navigation sites, Scarlett generally gets about 8/10 in terms of ease of use, performance, cost-effectiveness, interface, accuracy and other dimensions, and the overall score is about 8/10. The most recognized points among users are concentrated in three aspects: first, it can live directly in the team's existing Slack, and there is no new interface to learn; second, the integrated library is huge, and complex multi-step tasks can be completed in one go; third, the context understanding based on Claude is robust, and the configuration of recurring tasks is simple, which can significantly reduce manual chores.   The negative feedback is also concentrated: it requires the team itself to be active on Slack or iMessage, and the value is greatly reduced without these two scenarios; advanced workflows often require clear instructions to achieve good results, and the fault tolerance for vague requirements is limited; currently only supports text interaction, without voice or visual interfaces; the execution effect depends on the availability of third-party tool APIs, and the other party will get stuck if the other party is unstable.

• 6. Industry evaluation (918 chars)

  In Dir2AI's horizontal comparison, Scarlett's quality score is about 90.2 (top is 92.1, Claude Tag is 89.2, and Viktor is 89.2), and its average response time is about 25 seconds (the fastest is 9.8 seconds, and Viktor is 14 seconds). It falls into the range of "mid-to-high quality and steady speed" - for a tool that really does hands-on work rather than just pronouncing words, 25 seconds is a reasonable price.   Industry observers generally believe that Scarlett’s differentiation lies in “agent-like execution”: most AI assistants only generate text, while Scarlett maintains a persistent workspace, actually sends emails, updates CRM, and creates reports. It is natively integrated with Slack, credentials are hosted by a security agent, and the AI ​​model itself cannot see passwords and keys. This security design has been named and affirmed many times. Some people also pointed out that its initiative is still weak and is mainly driven by instructions issued by users. The independent orchestration of more complex scenes requires subsequent model capabilities to be completed.

• 7. Controversies and Risks (1155 chars)

  The risks Scarlett faces are mainly scene binding and compliance. Its capabilities are highly dependent on Slack/iMessage and are almost useless for teams that do not use these two platforms; at the same time, it lacks a visual drag-and-drop workflow editor, and operations teams accustomed to low-code construction may find it inflexible; for industries such as finance and medical care that are highly regulated and require localized deployment (on-prem), pure cloud solutions are difficult to directly meet.   In terms of data security, the official emphasized that each workspace has an isolated cloud infrastructure, and both transmission and static are encrypted. Data is not used for training models. Credentials are kept by a security agent with AES-256-GCM encryption and are only injected when actions are executed. Sensitive actions await manual approval. Scarlett can only see channels that have been explicitly invited to enter. This rhetoric reduces concerns, but as an early stage SaaS, its long-term compliance and audit capabilities still need to be evaluated by enterprise customers themselves. In addition, there is also a Chinese virtual companion app with the same name but completely unrelated "Scarlet AI" on the market. It is easy to be confused when searching. The official website domain name scarlett.ai / tryscarlett.ai should prevail.

• 8. Applicable groups and usage suggestions (800 chars)

  Scarlett is suitable for: remote teams that rely heavily on Slack, sales teams that need real-time synchronization of CRM, customer service teams that triage work orders, engineering teams that manage project flow, and small and medium-sized enterprises that do not have full-time automation engineers.   It is not suitable for: teams that do not use Slack/iMessage, teams that require visual drag-and-drop orchestration of complex processes, and industries with strong supervision that require local deployment. If your core requirement is to "automatically complete cross-system chores in a chat with just one sentence," Scarlett is a bucket-shaped choice worth trying; if what you need is general conversation or content creation, pure chat assistants such as ChatGPT and Claude are more lightweight; if you need enterprise-level low-code process orchestration, you can consider using mature solutions such as Zapier and Make together.

• 9. Summarize (267 chars)

  Scarlett put the concept of "AI colleagues" into Slack - with 3,000+ integrations and Claude-driven end-to-end execution, teams can run real cross-system workflows in one sentence; it may not be the fastest, but it has done a pretty solid job of embedding automation into daily chats and lowering the barrier to entry.

No in-depth report yet

• 

  Kax844

  —

  我们自己没自动化工程师，以前这些杂活要么堆人要么堆流程工具。Scarlett 等于把「一句话驱动跨系统」这件事门槛打到地板，对中小团队是实打实的杠杆。不需要会写代码，业务同学自己就能让它跑起一条流水线。

---

• 

  zaultViperGray

  —

  对比过 Claude Tag 和 Viktor，Scarlett 质量分 90.2 不算最高，但它是真去发邮件、更新 CRM、建看板的，那俩主要还在聊天框里。对「要结果不要答案」的团队，这点差异是关键。速度慢几秒能接受，毕竟它真在干活不是只打字。

---

• 

  Nathan_MartinezZ

  —

  接入太简单了，两分钟就跑起来了。

---

• 

  SusanHoward52026

  —

  实测下来「干真活」不是吹的——上周让它把合并的 PR 和 Linear issue 转成发布说明，出来的 changelog 直接能发，工程同学基本零改动。比我们自己手写快太多，对没有专职技术写作的小团队是真省人，等于白捡一个文档工程师。

---

• 

  Jacqueline.Russell168

  —

  50 美金一个月，小团队算下来还行。

---

• 

  4P6ZP

  —

  安全这块是我最在意的——每个工作空间隔离、凭据由代理托管、AI 本身看不到密钥、数据不拿去训练。我们财务也敢接了。不过敏感动作还是要它先问一句再发，这个确认环节挺必要，别嫌烦，真出过一次误发就知道值了。

---

• 

  MargaretTorres1687

  —

  会议前简报这个场景真香——自动把参会人、公司背景、最近邮件线程汇成一份，我进会前扫一眼就行。以前这种准备起码十分钟，现在零成本。建议做销售的都开，尤其见客户前那份 company context 救过我好几次场。

---

• 

  Met_aWave

  —

  每天早上的团队简报真香，省了我半小时。

---

• 

  JackThomas_2022892

  —

  最大爽点是活在我们本来就用的 Slack 里，不用再开一个后台。@一下就有结果，新同事也没学习成本。我们之前试过几个自动化平台，光是教人画流程图就劝退一半人，Scarlett 这种聊天式反而全员用得起来。

---

• 

  ASICantKrause

  —

  太香了。

---

• 

  Cheryl.Allen_Max

  —

  我们运营三个人，把周报、日报、合同巡检全交给它了。周一是自动经营报告，工作日九点自动项目状态摘要，Drive 一有新合同就 alert 到 Slack。前两周还盯着它干没干对，现在基本放手了，等于白捡半个运营助理，人力直接腾出来做内容。

---

• 

  PatrickSanchezZ

  —

  我们销售组用它扫 Gmail 没回复的对话，自动起草跟进邮件还加提醒，这一周跟进率明显上来了。之前这种活要么漏要么拖，现在每天定点扫一遍，重要的客户一个没漏，销售总监特意问了我们用了啥工具。

---

• 

  MargieHale

  —

  回不去了。

---

• 

  CRussell_Plus

  —

  客服分诊挺准，工单归类省心。

---

• 

  4o9tdsd8m1

  —

  模型按任务挑，聊天用 Opus、编码用 Claude 5.6，感觉比单一模型稳，复杂报告也不翻车。之前用纯单一模型的助手写带数据的周报，经常把数字算错，Scarlett 这种分任务调度明显更靠谱，长文逻辑也连贯。

---

• 

  Ronald_Evans_77878

  —

  高级工作流还是得把指令说清楚，模糊一句话它容易跑偏。我们后来养成了「先给明确输入输出」的习惯，比如「从 X 拉数据、按 Y 分组、输出成 Z 格式发到某频道」，跑出来的质量立刻稳了。把它当实习生带，别当魔法棒挥。

---

• 

  ABrownIII

  —

  响应有点慢，平均二十多秒。

---

• 

  ZAalv

  —

  依赖第三方 API，上周 Salesforce 接口抖了一下，它卡半天，算是被绑在别人稳定性上了。这种跨系统工具的通病，关键任务最好让它跑完再人工瞄一眼，别完全放手。

---

• 

  rptv2

  —

  文本交互没毛病，但不能语音、没有可视化拖拽编排，复杂流程只能靠自然语言描述，比 Zapier 那种画流程图还是费脑子。如果你们团队习惯了低代码拖拽，刚切过来会不适应，建议先用它跑标准化的高频任务。

---

• 

  MSullivan520

  —

  离了 Slack 基本废，我们外包团队用飞书，只能自己手动搬，体验割裂。

---

• 

  NicoleRoberts_Pro

  —

  适合重度 Slack 团队，飞书党慎入。

---

• 

  KatherinePerry_X

  —

  50 美元免费额度够试一阵，我们先用着看长期值不值。目前每天调用的 credits 还不低，月底得算算账。如果高频使用，25,000 credits 一个月对一个五人小团队其实不算宽裕，重活多的得留意用量。

No reviews yet