Fireflies 3

• 1. overview (723 chars)

  Fireflies.ai is an AI conference assistant focusing on "conference intelligence". The Fred robot automatically joins Zoom, Google Meet, Teams and other meetings, completes recording, transcription, summarization, action item extraction, and pushes conclusions to more than 100 work systems such as CRM, Slack, and project management tools. It has served more than 1 million enterprises and processed more than 2 billion meeting minutes. In June 2025, its valuation exceeded US$1 billion through a secondary offer and became profitable. This entry is named "Fireflies 3" in "Product Express". It is actually the latest capability collection of the same product (Live Assist, Desktop App, MCP Server, AskFred, Talk to Fireflies, etc. are all updated in 2025-2026). It is not an official single generation version number and is subject to manual review.

• 2. Product background (689 chars)

  Fireflies was founded in 2016 by Krish Ramineni (CEO, Indian-American, from the University of Pennsylvania, former Microsoft product manager) and Sam Udotong (CTO, Nigerian-American, MIT aerospace and computer background). The two tried six failed projects before they founded the company. In the early days, they even relied on the founders to pretend to be "AI robots" to manually type notes and deliver services to verify demand. Headquartered in San Francisco, the company has raised approximately $19.1 million in total funding, including a $14 million Series A in 2021 led by Khosla Ventures, and has not raised additional funding since then. Profitability will be achieved from 2023, with a secondary offer valuation of US$1 billion in 2025, and third-party estimates of ARR of approximately US$11 million.

• 3. Product functions and usage experience (1969 chars)

  At its core is a conference bot named Fred. It can automatically join a meeting according to calendar preferences, or be manually invited to a meeting via fred@fireflies.ai to complete audio and video recording, speaker recognition, transliteration and summary; it also supports Chrome extension (enter Google Meet without bot), desktop system-level recording, mobile terminal, and uploaded audio and video files (MP3/MP4/WAV/M4A) transcoding.   At the transcription level, the official claim is that the English accuracy rate is about 95%–99%, supports more than 100 languages, and includes multi-language automatic detection. However, third-party actual testing shows that: it is about 90%–95% in clean audio, native speakers, and low-overlapping scenarios. It drops to 60%–85% in heavy accents and noisy environments. Legal or compliance scenarios still require manual review.   On top of transcription, Fireflies offers three layers of capabilities. One is the AI ​​summary: overview, key points, action items, chapters divided by timestamps, and support for expanding single key points. The second is AskFred, a conversational AI that can ask questions for a single session or across all historical meetings (questions such as "When was the last time I discussed the budget with the client?" are returned in seconds). The third is 200+ "AI Skills" automated workflows, which can automatically generate follow-up emails, candidate scorecards, CRM updates, etc. after the meeting, covering sales, recruitment, marketing, R&D, finance and other scenarios.   New actions in 2026 will push the product from "recording" to "real-time collaboration." Live Assist will be launched in May 2026, and can give real-time suggestions, coaching and answers in meetings; Desktop App will be released at the same time, supporting system-level audio capture outside the browser; MCP Server will be expanded to 17 core tools in early 2026, allowing agents such as Claude, GPT, Cursor, etc. to directly query your meeting data; "Talk to Fireflies" is connected to Perplexity real-time network search, supporting "Hey Fireflies" voice or "/ff" command to answer questions instantly in meetings. There’s also conversational intelligence (speaker duration, sentiment analysis, topic tracking) and Soundbites audio clip editing.

• 4. Pricing and business model (991 chars)

  Official pricing is based on annual monthly fees: Free forever (400 minutes of team storage, 20 AI credits, 2 hours of single-session recording limit, limited AskFred); Pro $10 per seat (8000 minutes of storage, 20 credits, 3 hours of recording, including video recording, AI Skills, Voice Agents, unlimited integrations); Business $19 per seat (unlimited storage, 30 credits, including multi-language mode, conversational intelligence, team analytics, public meetings); Enterprise per seat $39 and only paid annually (50 credits, SSO/SCIM, audit logs, HIPAA, private storage, custom retention rules).   The key change is the introduction of the "AI credits" credit system in 2026: the basic subscription only covers "basic" capabilities such as transcription and summarization. Advanced capabilities such as AskFred, AI Skills, and Voice Agents are consumed by credits. Pro users only have 20 points per month, and heavy users will reach the top within a few days. Additional credit packages range from 5 to 600 US dollars. This has led to criticism of a “subscription doll” approach – a low entry price of $10 that can easily turn into a $40 monthly fee after frequent use.

• 5. User reviews (836 chars)

  Positive feedback focused on time savings and searchability. G2 has a score of about 4.8/5. Users generally praise its high transcription accuracy, integrated ecosystem across Zoom/Meet/Slack/CRM, and meeting search that can “find the original sentence from a few months ago in one sentence.” The sales team especially appreciates the automatic backfilling of the CRM, reducing manual entry after meetings.   The negative points are concentrated in three areas: First, the learning curve is steep, there are too many functions, and new users need to adjust parameters before they can use it properly; second, there is anxiety about credit consumption, and heavy users complain that advanced functions are "stuck"; third, the intrusion of the visible bot - the robot enters the meeting and clearly states that it is "recording", which will make some customers more vigilant. There are also compliance risks in dual-consent jurisdictions such as California, Illinois or the European Union.

• 6. Industry evaluation (547 chars)

  In a horizontal comparison of AI conference assistants in 2026, Fireflies is often positioned as "the first choice for CRM and production flow teams." Compared with Otter.ai, Fireflies has superior integration breadth (50+ tools) and analysis capabilities (speaker duration, emotion, theme), and its transcription accuracy is slightly lower (about 91% vs 94%); compared with Fathom, Fathom is free and has a clean experience on Zoom, but has strong platform bindings and shallow integration. The media generally recognizes its “agentic layer” positioning—not just a chatbot, but an intelligent agent that follows up on its own in the workflow.

• 7. Controversies and Risks (949 chars)

  The most prominent disputes arise over privacy and compliance. Competitor Otter.ai is facing the Brewer v. Otter.ai class action lawsuit, accusing it of recording and using it to train models without the consent of all attendees; Fireflies uses "no customer data to train models, users have data ownership" as a differentiating selling point, and provides SOC 2 Type II, GDPR, HIPAA (Enterprise Edition) and zero data retention commitments. However, visible bots still pose risks in dual-consent jurisdictions, and officials recommend that sensitive jurisdictions must keep bots visible to indicate consent.   Secondly, there is confusion in naming and ability caliber: external names such as "Fireflies 3" are not official generations, and official abilities are continuously superimposed, which can easily confuse new users in version recognition. In terms of technical risks, the transcription accuracy drops significantly in scenarios with heavy accents, noisy, and rapid cross-speaking, and compliance records still require manual proofreading; the UI is biased toward novices and heavy use by individuals.

• 8. Applicable groups and usage suggestions (604 chars)

  It is suitable for teams that are in meetings every day and whose workflow endpoint is a CRM or project management system - sales, customer success, recruitment, and R&D reviews all like this set of automatic backfill. If you mainly use Zoom and have zero budget, Fathom’s free file is more cost-effective; if the accuracy of transcription is your top priority (discussing legal and technical architecture), Otter.ai is slightly better. It is recommended when using Fireflies: Set which calendars are monitored, which meetings are automatically joined, and robot visibility from the beginning, and evaluate the team's average monthly AI credit consumption to avoid frequent purchases after starting at a low price.

• 9. Summarize (256 chars)

  Fireflies.ai has turned "meeting minutes" into "meeting agents" and will stand in the first echelon by 2026 with its integration breadth, profit model and zero training data commitment; what really needs to be weighed is the hidden cost of the credit system and the compliance boundary of the visible bot.

No in-depth report yet

• 

  JanetMorgan

  —

  定价偏贵，但集成广度确实没对手能打。

---

• 

  Lydia_Wood369

  —

  免费的 800 分钟很快就满了，轻度用还行，天天开会真的不够，得往上升级。

---

• 

  KarenBrooksII

  —

  回不去了，开会再也不手敲笔记了。

---

• 

  RaymondPrice

  —

  AskFred 是真的好用，直接问「这周销售电话里客户主要的 objections 是什么」就能拉出来，不用再翻十几份转录稿。

---

• 

  ScottPatel

  —

  我们销售团队每次客户电话结束，它会自动把纪要和 action items 写进 HubSpot，省了每天一两小时手工录入。但 Pro 版的集成就被阉割了，HubSpot 和 Salesforce 推送都得升到 Business 才行，这点挺坑的，等于变相逼你买贵的档。

---

• 

  朱云

  —

  那个 notetaker bot 每次都坐进会议室，客户看到有点尴尬，敏感会议我都手动踢掉。

---

• 

  Teresa_CollinsZ

  —

  AI credits 这套真的要吐槽。Pro 一个月才 20 点，AskFred 和 AI Skills 一用就扣，重度用户月中就见底，再加购动不动就是 5 到 600 刀的包。10 刀的入门价看着香，真用起来轻松变 40 刀月费，订阅套娃实锤了。

---

• 

  Gloria_Bennett_Plus

  —

  转录准确率在干净音频下确实能到 90% 以上，但一遇到重口音和多人抢话就掉到七成，合规记录还是得人工校一遍。

---

• 

  任妍萱

  —

  界面只有英文，全球团队用着有点割裂。

---

• 

  Evelynbarsen

  —

  我们招聘团队用 Fireflies 筛候选人电话，可搜索转录对回溯太香了。不过有个坑：超过四五个人同时说话时，说话人识别经常张冠李戴，活儿得返工。另外客服响应是真慢，工单转半天没人回，有一次重要会议它没进还丢了记录。

---

• 

  浮生_5

  —

  免费档对比 Otter 的 300 分钟算大方了，先试试不亏。

---

• 

  MsRamsesBahena_dev

  —

  Live Assist 五月刚上，会议里能实时给提示和辅导，方向是对的，但还太早，复杂任务偶尔掉链子。

No reviews yet