Is Otter.ai: Record & Transcribe Meetings - Google Meet & Web Audio safe?
Otter.ai is low risk. After a signed-in user starts recording, it opens a recording tab, captures audio, opens Otter.ai's transcription WebSocket, sends a start message, then forwards audio buffers. Unauthenticated checks didn't trigger this; login is required.…
Who publishes itOtter.ai - no other listings under this identity, 3 shared hostnames
Otter.ai - no other listings under this identity, 3 shared hostnames
Shared hosts - 3 hostnames
Hostnames hardcoded in this extension that few other listings call. That can mean one operator behind both, and it can equally mean a small shared vendor, so it is context rather than a conclusion. Hosts that many listings call are left out: they are services, not connections.
Otter.ai findings
Tab audio streams to Otter.ai after recording starts
After a signed-in user starts recording, it opens a recording tab, captures audio, opens Otter.ai's transcription WebSocket, sends a start message, then forwards audio buffers.
Unauthenticated checks didn't trigger this; login is required.
What actually happens
You start an Otter.ai recording from the extension.
The code path requires an authenticated Otter.ai session and a recording action.
The extension captures audio from the active browser tab and sends it to Otter.ai for transcription.
The recording tab sends a JSON start message first, followed by binary audio buffers.
Fields and buffers sent by the recording flow
- Recording identifierspeech_id from the recording-start response
Lets Otter.ai associate the audio stream with the transcript you started.
- Processed offsetoffset: 0
Tracks how much of the audio stream has already been acknowledged.
- Browser-tab audio frames16 kHz audio buffer, 1024 samples
Audio playing in the captured tab can include meeting speech, media playback, or other spoken content.
- Capture stream handlestream handle for the active tab audio
Connects the recording tab to the browser tab whose audio is being recorded.
Captured request
WebSocket upgrade for transcription; available traffic evidence did not record a request body for this connection.
The code that does this
Shipped and deobfuscated recording path evidence
Record-button handler passes the server-provided WebSocket URL and speech ID
src/content-script/contentscript.jsfunction EV() { const [, t] = gt(So), [, e] = gt(Q1), [, n] = gt(a6), [r] = gt(o6), [i] = gt(N$), { mutate: o } = sGe(), { data: s } = Co(), a = l6(sv.audioTag, "enabled"), l = et() ? yve : _ve; return { startRecording: h.useCallback(async (d = !1) => { if (!await mZe()) { Is(), Os(c.jsx(Ls, { "data-testid": "cannot-start-recording-toast", description: JD, altText: JD, showCloseIcon: !0 }), { duration: Ss.endCurrentRecording }); return } s && o({ queryParams: { appid: l, uuid: gE(), userid: JSON.stringify(s.userid), start_time: Math.floor(Date.now() / 1e3), ignore_event: !0 } }, { onSuccess: p => { if (et() && a) { pZe(p.ws_url, p.speech_id, d, r, i, JSON.stringify(s.userid), p.otid, l), e(p.otid), t("liveTranscription"), n(!0); return } const g = { action: Pi.BG_RECORDING_ACTION, type: "start", wsUrl: p.ws_url, speechId: p.speech_id, micEnabled: d, micMuted: r, deviceLabel: i, userId: JSON.stringify(s.userid), otid: p.otid, appId: l }; Rn.runtime.sendMessage(g), e(p.otid), t("liveTranscription") }, onError: p => { const g = p; g.message && (Is(), Os(c.jsx(Ls, { "data-testid": "recording-error-toast", description: g.message, altText: g.message, showCloseIcon: !0 }), { duration: Ss.error })) } }) }, [r, i, a, e, t, n, o, s, l]) } }Background script obtains a tabCapture stream ID and starts the recording tab
src/background-script/background.jsasync function Nd(t) { if (!t.tabId) throw new Error("[Background] No active tab ID when starting recording"); let e; try { e = await chrome.tabs.get(t.tabId) } catch (b) { throw new Error(`[Background] Active tab does not exist anymore, ${b}`) } const r = await navigator.permissions.query({ name: "microphone" }), a = r.state === "prompt" || r.state === "denied", o = t.micEnabled && a, s = new Promise(b => { const P = y => { y.action === yt.RT_READY && (chrome.runtime.onMessage.removeListener(P), b(!0)) }; chrome.runtime.onMessage.addListener(P) }), p = await new Promise((b, P) => { chrome.tabs.create({ url: chrome.runtime.getURL("recordingtab.html"), pinned: !0, active: o }, y => { if (chrome.runtime.lastError) return P(new Error(`[Background] ${chrome.runtime.lastError.message}`)); b(y.id) }) }); if (!p) { Rr({ errorMessage: "[Background] Failed to create recording tab", shouldThrow: !0, csTabId: e.id }); return } await s; const d = { action: yt.RT_MIC_ENABLED, type: t.micEnabled ? "enabled" : "disabled" }; chrome.tabs.sendMessage(p, d), Di(!0, p, e.id), await new Promise(b => { setTimeout(() => { b(!0) }, 1e3) }); const m = await new Promise((b, P) => { chrome.tabCapture.getMediaStreamId({ targetTabId: e.id, consumerTabId: p }, y => { if (chrome.runtime.lastError) { P(new Error(`[Background] Tabcapture: ${chrome.runtime.lastError.message}`)); return } b(y) }) }), g = { action: yt.RT_RECORDING_ACTION, type: "start", data: { wsUrl: t.wsUrl, speechId: t.speechId, tabAudioStreamId: m, micEnabled: t.micEnabled, micMuted: t.micMuted, deviceLabel: t.deviceLabel, tabFocused: o, userId: t.userId, otid: t.otid, appId: t.appId } }; chrome.runtime.sendMessage(g) }Recording tab opens the WebSocket, sends the start message, then sends binary audio buffers
src/recordingTab/index.jsasync function Cp(e) { if (pe = new Wr(e.wsUrl), Fr = e.speechId, yg = e.micEnabled, Fi = e.userId, Ii = e.otid, Pi = e.appId, ef = Math.floor(Date.now() / 1e3), Wi = new Op, pe.onopen = async () => { pe == null || pe.send(JSON.stringify({ action: "start", speech_id: e.speechId, offset: ru })), fu = !0, await Wi.read(e.speechId, a => { if (a.length > 0) for (const l of a) pe == null || pe.send(l) }) }, pe.onmessage = async a => { const l = Np(a.data); if (!l) return; const n = hp.safeParse(l); if (!n.success) return; n.data.type === "ack" && (ru = n.data.result.processed_offset + 1, ru > Ir && (await Wi.remove(e.speechId, 2 * (ru - Ir)), Ir = ru)) }, pe.addEventListener("error", () => { fu = !1 }), pe.addEventListener("close", () => { fu = !1 }), e.micEnabled) { e.tabFocused && await navigator.mediaDevices.getUserMedia({ audio: !0 }); const l = (await navigator.mediaDevices.enumerateDevices()).filter(u => u.label === e.deviceLabel && u.kind === "audioinput"); let n = "default"; l.length > 0 && l[0] && (n = l[0].deviceId), jt = await navigator.mediaDevices.getUserMedia({ audio: { deviceId: { exact: n } } }) } const t = await navigator.mediaDevices.getUserMedia({ audio: { mandatory: { chromeMediaSource: "tab", chromeMediaSourceId: e.tabAudioStreamId } } }); e.micMuted && tf(!0), Oe && await Oe.close(), Oe = new AudioContext({ latencyHint: "interactive", sampleRate: 16e3 }), Oe.state === "suspended" && (console.error("[Recording tab] AudioContext is suspended on creation"), await Oe.resume()); try { await Oe.audioWorklet.addModule("src/shared/tab-audio-processor.js") } catch (a) { throw new Error(`[Recording tab] ${a}`) } su = Oe.createMediaStreamSource(t), jt && (rl = Oe.createMediaStreamSource(jt)), su.connect(Oe.destination), xt = pg(rl, su, Oe), cu = new AudioWorkletNode(Oe, "tab-audio-processor"), xt.connect(cu), cu.port.onmessage = async a => { const l = a.data.audioBuffer.buffer; fu && (pe == null || pe.send(l)), await Wi.append(e.speechId, l) } }Audio worklet converts page audio into Int16 buffers posted to the recording tab
src/shared/tab-audio-processor.jsvar _ = function(t) { "use strict"; var l = Object.defineProperty; var b = (t, e, r) => e in t ? l(t, e, { enumerable: !0, configurable: !0, writable: !0, value: r }) : t[e] = r; var f = (t, e, r) => b(t, typeof e != "symbol" ? e + "" : e, r); class e extends AudioWorkletProcessor { constructor() { super(...arguments); f(this, "bufferSize", 1024); f(this, "buffer", new Int16Array(this.bufferSize)); f(this, "bytesWritten", 0) } isBufferFull() { return this.bytesWritten === this.bufferSize } appendToBuffer(s) { this.isBufferFull() && this.flush(), this.buffer[this.bytesWritten] = s, this.bytesWritten += 1 } flush() { let s = this.buffer; this.bytesWritten < this.bufferSize && (s = s.slice(0, this.bytesWritten)), this.port.postMessage({ eventType: "data", audioBuffer: s }), this.bytesWritten = 0 } process(s) { const a = s.length; for (let u = 0; u < a; u++) { const i = s[u]; if (!(!i || !i[0])) for (let o = 0; o < i[0].length; o++) { if (!i.length) continue; const n = Math.max(-1, Math.min(1, 1 * i[0][o])), h = n < 0 ? n * 32768 : n * 32767; this.appendToBuffer(h) } } return !0 } } return registerProcessor("tab-audio-processor", e), t.TabAudioProcessor = e, Object.defineProperty(t, Symbol.toStringTag, { value: "Module" }), t}({});Remote service receiving the recording stream
- ws.aisense.com
Otter.ai transcription WebSocket that receives the start message and browser-tab audio frames.
Greenhouse candidate name sent to Otter search
With Greenhouse integration on, clicking Otter's control on an interview page reads the candidate name, page URL, job title, and form context, then queries Otter's advanced_search endpoint for matches.
No live request captured.
What actually happens
You use Otter's Greenhouse integration on an interview feedback page.
The content script reads candidate and form context, then searches Otter conversations using the candidate name.
Greenhouse page data used by the integration
- Candidate nameJane Candidate (illustrative)
This identifies the applicant whose Greenhouse page you are viewing.
- Greenhouse page URLhttps://app.greenhouse.io/guides/interview_kits/12345/people/67890 (illustrative)
This ties the action to a specific Greenhouse workflow page in your browser.
- Interview form fieldsProblem Solving, Communication, Technical Depth (illustrative)
This describes the evaluation fields visible on the page.
- Job and interview metadataSenior Product Manager, Onsite Interview (illustrative)
This adds hiring context to the transcript search and generated feedback.
Captured request
The source builds this GET pattern from the candidate name; no live Greenhouse response was recorded in the supplied evidence.
The code that does this
Greenhouse content script extracts candidate/page context and sends the search message
Candidate-name extractor
src/prototypes/greenhouse/content-script.jsasync function Cy() { var e, a; const t = await $t.candidateName.getSelectorString(); return ((a = (e = document.querySelector(t)) == null ? void 0 : e.textContent) == null ? void 0 : a.trim()) || void 0 }Greenhouse page-context assembler
src/prototypes/greenhouse/content-script.jsasync function Ry() { const [t, e, a, l, r, n, o, i, c] = await Promise.all([_y(), Ay(), ky(), Ty(), Dy(), Fy(), Ny(), zy(), Cy()]), u = { formFields: e, pageUrl: window.location.href, extractedAt: new Date().toISOString() }; if (t !== "unknown" && (u.pageType = t), i !== void 0 && (u.jobTitle = i), c !== void 0 && (u.candidateName = c), l !== void 0 || a.length > 0 || r !== void 0 || n.length > 0 || o !== void 0) { const g = {}; l !== void 0 && (g.interviewStep = l), a.length > 0 && (g.focusAttributes = a), r !== void 0 && (g.interviewGuide = r), n.length > 0 && (g.scorecardAttributes = n), o !== void 0 && (g.keyTakeawaysPrompt = o), u.metadata = g } return u }User-click handler
src/prototypes/greenhouse/content-script.jsasync function jy() { var r, n; if (!await My()) { console.warn("[Greenhouse Prototype] Not an interview feedback page, skipping"), alert("This doesn't appear to be a Greenhouse interview feedback page."); return } const t = await Ry(); console.info("[Greenhouse Prototype] Extracted data:", t); const e = t.formFields.length > 0, a = !!((n = (r = t.metadata) == null ? void 0 : r.scorecardAttributes) != null && n.length); if (!e && !a) { console.warn("[Greenhouse Prototype] No form fields or scorecard attributes found"), alert("No form fields or scorecard attributes found on this page. The page structure may have changed."); return } if (!t.candidateName) { console.warn("[Greenhouse Prototype] No candidate name found"), alert("Cannot find candidate name on this page. Please make sure you are on a Greenhouse interview feedback page."); return } const l = await Uy(t.candidateName); l && await By(l, t) }Search message sender
src/prototypes/greenhouse/content-script.jsasync function Uy(t) { console.info(`[Greenhouse Prototype] Searching for conversations with candidate: ${t}`); try { const e = { type: "GREENHOUSE_SEARCH_CONVERSATIONS", candidateName: t }, { success: a, error: l, conversations: r = [] } = await chrome.runtime.sendMessage(e); if (!a) return console.error("[Greenhouse Prototype] Failed to search conversations:", l), alert(`Failed to search conversations: ${l}`), null; if (console.info("[Greenhouse Prototype] Found conversations:", r), r.length === 0) return alert(`No interview conversations found for "${t}". Please make sure you have a recorded interview with this candidate.`), null; if (r.length === 1) { const { speech_otid: o, title: i } = r[0]; return console.info(`[Greenhouse Prototype] Auto-selected conversation with speech_otid=${o} and title=${i}`), o } const n = await Oy(r); return n ? n.speech_otid : (console.info("[Greenhouse Prototype] User cancelled conversation selection"), null) } catch (e) { return console.error("[Greenhouse Prototype] Uncaught error in findSpeechOtidByCandidateName:", e), alert("Failed to find conversation. Please try again."), null } }The code that does this
Background service worker converts the candidate name into an Otter search request
Advanced-search fetch helper
src/background-script/background.jsasync function yd(t) { const e = `${rn}/advanced_search?query=${encodeURIComponent(t)}&relevance=true&size=5`, n = await de(Nt, "csrftoken"); return await fetch(e, { method: "GET", credentials: "include", headers: { "x-csrftoken": n } }) }Search response handler
src/background-script/background.jsasync function Pd(t) { try { console.info("[Greenhouse Prototype] Searching for conversations with candidate:", t); const e = await yd(t); if (console.info("[Greenhouse Prototype] Search response status:", e.status), !e.ok) { const r = await e.text(); return console.error("[Greenhouse Prototype] Search API error:", r), { success: !1, error: `Search API returned ${e.status}: ${r}` } } const n = await e.json(); return console.info("[Greenhouse Prototype] Search results:", n), console.info("[Greenhouse Prototype] Conversations found:", n.hits.length), { success: !0, conversations: n.hits } } catch (e) { return console.error("[Greenhouse Prototype] Search exception:", e), { success: !1, error: e instanceof Error ? e.message : String(e) } } }Greenhouse message dispatcher
src/background-script/background.jsasync function Id(t) { if (t.type === "GREENHOUSE_SEARCH_CONVERSATIONS") { console.info("[Greenhouse Prototype] Searching for conversations"); const e = await Pd(t.candidateName); return console.info("[Greenhouse Prototype] searchConversations result:", e), e } if (t.type === "GREENHOUSE_GENERATE_FORM_FILLS") { console.info("[Greenhouse Prototype] Generating form fills"); const e = await wd(t.speechOtid, t.payload); return console.info("[Greenhouse Prototype] generateGreenhouseFormFills result:", e), e } return { success: !1, error: "Unknown message type" } }Remote service receiving the Greenhouse search query
- otter.ai
Receives the candidate-name query at /forward/api/v1/advanced_search.
Gmail compose context sent to Otter LLM proxy
Using Otter's email controls in Gmail reads the sender account tag and document title, combines these with recipients, meeting title, and draft text, then posts the prompt to Otter's LLM proxy.
No live Gmail request captured; path exists.
What actually happens
You ask Otter to generate or refine an email in Gmail.
The extension gathers Gmail compose context and sends a text prompt to Otter's LLM proxy.
Gmail context placed into the LLM prompt
- Sender account contextalex@acme.example from Inbox - Gmail (illustrative)
This can identify which Gmail account or organization is being used for the draft.
- Recipient emailssam@vendor.example, priya@acme.example (illustrative)
This identifies the people or organizations involved in the drafted message.
- Email bodyThanks for meeting today. Here are the pricing questions we discussed. (illustrative)
This is the draft text that the service uses to generate or refine the message.
- Meeting titleProject Sync follow-up (illustrative)
This adds business context to the generated subject line or follow-up email.
Captured request
The source posts the Gmail prompt to this endpoint; no live Gmail request body was recorded in the supplied evidence.
The code that does this
Gmail content script reads page context and requests subject generation
Gmail compose-generation flow
src/prototypes/gmail/content-script.jsasync function fae(t, e, n, r) { var a; try { let i = aae(e); if (r && r.length > 0) { const d = sae(r); i = `${i}${d}`, console.info(`[Email Generator] Added ${r.length} thread emails to prompt`) } const o = { type: "GMAIL_GENERATE_FOLLOWUP_EMAIL", speechOtid: t.id, prompt: i }, s = await chrome.runtime.sendMessage(o); if (!s.success) throw new Error(s.error || "Failed to generate email"); const l = s.data, c = { subject: l.subject || "Follow-up", recipients: l.recipients || [], body: l.generated_draft_email || "" }; if (console.info("[Email Generator] Generated email:"), console.info(" - Subject:", c.subject), console.info(" - Recipients:", c.recipients), console.info(" - Body length:", c.body.length), c.body) { console.info("[Email Generator] Refining tone and generating subject..."); const d = chrome.runtime.sendMessage({ type: "GMAIL_REFINE_EMAIL_TONE", emailHtml: c.body }), f = await Me.gmail.senderEmailMeta.getSelectorString(), m = `${((a=document.querySelector(f))==null?void 0:a.getAttribute("content"))??""} ${document.title}`, b = chrome.runtime.sendMessage({ type: "GMAIL_GENERATE_SUBJECT", meetingTitle: t.name, emailBody: c.body, recipients: c.recipients, senderContext: m }), [x, _] = await Promise.all([d, b]); x != null && x.success && x.data ? (c.body = x.data, console.info("[Email Generator] Tone refinement applied, new body length:", c.body.length)) : console.warn("[Email Generator] Tone refinement failed, using original:", x == null ? void 0 : x.error), _ != null && _.success && _.data ? (c.subject = _.data, console.info("[Email Generator] Subject generated:", c.subject)) : console.warn("[Email Generator] Subject generation failed, using original:", _ == null ? void 0 : _.error) } if (!await lae(c, n)) throw console.error("[Email Generator] Failed to insert content into Gmail"), new Error("Failed to insert content into Gmail"); return console.info("[Email Generator] Successfully generated and inserted email"), !0 } catch (i) { console.error("[Email Generator] Error in generateAndInsertEmail:", i), console.error("[Email Generator] Error stack:", i instanceof Error ? i.stack : "No stack"); const o = i instanceof Error ? i.message : String(i); return alert(`Error generating email: ${o}`), !1 } }The code that does this
Background service worker posts the Gmail prompt to Otter
LLM proxy POST helper
src/background-script/background.jsasync function Io(t, e = {}) { var n; try { const r = "https://otter.ai", a = await de(r, "csrftoken"), o = `${r}/forward/api/v1/get_llm_proxy_response`, s = { llm_request: { chat_content: [{ type: "text", text: t }], sys_content: e.systemPrompt ? [{ type: "text", text: e.systemPrompt }] : [], option: { models: [e.model ?? ta.model], temperature: e.temperature ?? ta.temperature, max_tokens: e.maxTokens ?? ta.maxTokens } } }, p = await fetch(o, { method: "POST", credentials: "include", headers: { "x-csrftoken": a, "Content-Type": "application/json" }, body: JSON.stringify(s) }); if (!p.ok) { const m = await p.text(); return { success: !1, error: `LLM proxy API returned ${p.status}: ${m}` } } const d = await p.json(); return d.status !== "OK" || !((n = d.response) != null && n.content) ? { success: !1, error: d.message || "LLM proxy returned no content" } : { success: !0, content: d.response.content } } catch (r) { return { success: !1, error: r instanceof Error ? r.message : String(r) } } }Gmail subject prompt builder
src/background-script/background.jsasync function vh(t, e, n, r) { console.info("[Gmail Prototype] Generating subject via LLM proxy"); const a = ['Generate a short email subject line in the format "My Company Name <> Their Company Name".', "Infer my company name from a combination of Gmail page elements and document.title:", `<my-context>${r}</my-context>`, "Infer their company name from the following:", `<recipient-emails>${n.join(", ")}</recipient-emails>`, `<meeting-title>${t}</meeting-title>`, `<email-body>${e}</email-body>`, `Return ONLY the subject line text, nothing else. If you can't infer both, return "Follow-up"`].join(``), o = await Io(a, { systemPrompt: "You generate concise email subject lines. You return only the subject line text, no quotes, no explanation.", maxTokens: 100 }); return o.success ? { success: !0, ...o.content !== void 0 && { subject: o.content } } : (console.error("[Gmail Prototype] Generate subject failed:", o.error), { success: !1, ...o.error !== void 0 && { error: o.error } }) }Remote service receiving the Gmail prompt
- otter.ai
Receives the Gmail compose prompt at /forward/api/v1/get_llm_proxy_response.
+3 more findings not shown
Otter.ai version history
3.8.2
Mar 4, 2026
+4 more
3.10.1
Aug 7, 2026
3.11.0
Oct 3, 2026
3.12.0
Nov 11, 2026
Chrome Web Store listing
- Versions
- 3.12.0 latest of 8 seen since March 4, 2026
- Store rating
- 4.1 from 143 ratings
- Last updated
- July 7, 2026
- Size
- 1.59 MiB
- Privacy policy
- otter.ai
