Quick start
Audora has three moving parts: the bot that joins your calls, the pipeline that processes the transcript and generates outputs, and the integrations that push those outputs into your tools. You do not need to configure all three before using it - the bot and the outputs work immediately. Integrations can be set up at any time.
Before your first call
Create your account
Sign up at audora.agency. You are automatically placed on a 14-day Studio plan trial. No credit card required to start.
Connect your calendar (recommended) or use direct links
Go to Settings → Calendar. Connect Google Calendar via OAuth. Audora will detect upcoming meetings and ask per-call whether you want the bot to join. If you prefer not to connect your calendar, you can paste any Zoom, Meet, or Teams link manually from the dashboard when needed.
Connect at least one build tool integration
Go to Settings → Integrations. Connect Cursor (requires GitHub OAuth), Linear (OAuth), or Notion (OAuth). Lovable and Bolt require no connection - links are generated automatically. You can skip this and add integrations after your first call.
Running your first call
Invite the bot
Either confirm the bot from the calendar prompt, or paste your meeting link into the dashboard and click Send bot. The bot joins at the meeting's scheduled start time, or immediately if the meeting is already in progress.
Run your client call normally
The bot joins as a participant named "Audora" and is visible in the participant list. It does not interact with the call. You do not need to do anything differently - just have your conversation.
End the call
Processing starts the moment the meeting ends. You will receive an email notification when all six outputs are ready - typically within 60 seconds for calls under 90 minutes.
Review outputs and push
Open the dashboard. All six outputs are available. Review them, edit if needed, then push to your connected tools with one click per integration.
Tip: Run Audora on a real client call first, not a test call with yourself. Output quality is calibrated for actual client conversations - discovery calls, requirement sessions, project kickoffs. A two-person test call with no real content will produce outputs that do not reflect what the product can actually do.
How Audora works
Understanding what Audora does at each stage helps you use it more effectively and debug problems when they occur.
Stage 1 - The bot joins
When you invite the bot to a call, Audora sends a software participant to your meeting via the platform's join API. This participant appears in the meeting exactly like a human attendee - they show up in the participant list, they can be muted or removed by the host, and they are subject to the same waiting room and admission rules.
The bot captures audio from the call in real time. It does not capture video. It identifies each speaker separately using voice diarization - the process of attributing audio segments to individual speakers - and produces a speaker-labeled audio stream.
Stage 2 - Transcription
As audio is captured, it is transcribed in real time using a speech-to-text model. The transcript is built line by line, attributed to each speaker as the call progresses. The transcription is finalized immediately when the call ends.
Transcription accuracy depends on audio quality. Calls with clear audio, minimal background noise, and speakers who are not talking over each other produce accurate transcripts. Calls with heavy accents, technical jargon specific to a narrow field, or poor microphone quality may have lower accuracy in places.
Stage 3 - Output generation
Once the final transcript is ready, Audora runs it through three sequential AI processing steps:
- Extraction - raw facts are pulled from the transcript: what was requested, what was agreed to, who said what, what constraints were mentioned, what is explicitly out of scope.
- Structuring - extracted information is organized into the six output formats. This step determines what goes into the requirements document versus the PRD versus the task breakdown.
- Client document generation - the SOW and client brief are written in a professional, client-facing tone based on the structured outputs from step two.
Three focused steps produce significantly better outputs than one large prompt. It reduces hallucination, improves formatting consistency, and makes debugging easier when something looks wrong.
Stage 4 - Push to integrations
Outputs are pushed to connected integrations automatically after processing. You can also trigger pushes manually from the dashboard per output, per integration. Edits made in the dashboard before pushing are included in the pushed version.
What Audora does not do
- It does not summarize loosely - if something was not said in the call, it is not in the outputs
- It does not make judgment calls about what to build - it documents what was discussed
- It does not interact with call participants in any way
- It does not retain audio after transcription is complete
- It does not use your data to train AI models
Creating your account
Signing up
Go to audora.agency and click Get started. Enter your email and create a password of at least 12 characters. You will receive a verification email - click the link to activate your account. You are automatically placed on a 14-day Studio trial.
Use your work email. If you are on a Studio or Agency plan with team members, everyone on the team needs to be able to verify their email address. Use a domain you control.
Account settings
Settings is accessible from the top-right menu in the dashboard. It is organized into the following sections:
| Setting | What it controls |
|---|---|
| Profile | Your name, email address, and password. Email changes require re-verification. |
| Calendar | Google Calendar connection. Controls whether Audora can detect upcoming meetings. |
| Integrations | Connected build and project management tools. Authorize, deauthorize, and configure push behavior per integration. |
| Notifications | Email notification preferences. You can control which events trigger an email: outputs ready, call limit warnings, processing failures. |
| Team | Invite and manage team members. Available on Studio and Agency plans. |
| Billing | Subscription plan, payment method, invoice history, and cancellation. |
| Security | Active sessions, session revocation, and password change. |
Changing your email address
Go to Settings → Profile → Change email. Enter the new email address. A verification link is sent to the new address. Your email does not change until you click the link. Your old email address receives a notification that a change was requested - if you did not request it, contact support immediately.
Changing your password
Go to Settings → Profile → Change password. You must enter your current password before setting a new one. Changing your password revokes all active sessions except your current one - any other devices or browsers where you are logged in will be signed out.
Deleting your account
Go to Settings → Billing → Delete account. Account deletion is permanent and immediate. All data - transcripts, outputs, call history - is permanently deleted. Active subscriptions are cancelled immediately with no refund for the remaining billing period unless you are within the 7-day refund window. See the Refund Policy.
Setting up the bot
Method 1 - Calendar connection (recommended)
Connecting Google Calendar is the easiest way to use Audora. Once connected, Audora monitors your calendar for upcoming meetings. Before each meeting, you receive a prompt in the dashboard (and optionally by email) asking whether you want the bot to join. You confirm or decline per call - Audora never joins without your explicit per-call confirmation.
Connecting Google Calendar
Go to Settings → Calendar
Click Connect Google Calendar. You are redirected to Google's OAuth consent screen.
Authorize Audora
Grant Audora read access to your calendar events. Audora reads event titles, times, and meeting links. It does not modify your calendar, create events, or read event descriptions or attendee email addresses.
Review upcoming meetings
After connecting, the dashboard shows a list of upcoming meetings detected from your calendar. Toggle on the ones you want the bot to attend. You can change this up to 10 minutes before the meeting starts.
What Audora reads from your calendar: event name, start time, end time, and the video conference link embedded in the event (Zoom link, Google Meet URL, or Teams link). It reads nothing else. Event descriptions, attendee lists, and organizer details are not accessed.
Method 2 - Manual invite
If you prefer not to connect your calendar, or if you want to invite the bot to a meeting not on your calendar, use the manual invite from the dashboard.
Open the dashboard
Click New call in the top right.
Paste the meeting link
Paste a Zoom, Google Meet, or Microsoft Teams link. Audora validates the link format before accepting it.
Set a time (optional)
If the meeting has not started yet, you can specify when the bot should join. If you leave this blank, the bot attempts to join immediately. If the meeting has not started, the bot will wait in the waiting room for up to 15 minutes.
Send the bot
Click Send bot. You receive a confirmation. The bot's status is shown in the dashboard in real time.
Cancelling a bot invite
From the dashboard, find the pending call and click Cancel. The bot will not join. If the bot has already joined, clicking Cancel will remove it from the meeting - processing will run on whatever audio was captured up to that point.
Bot join limits
The bot will wait in a meeting's waiting room for up to 15 minutes before timing out. If the meeting uses a waiting room and the host does not admit the bot within 15 minutes, the bot exits and you are notified. The call is not counted against your monthly limit in this case.
Bot behavior during calls
What the bot does
- Appears in the participant list as "Audora"
- Captures audio from all participants throughout the call
- Performs real-time speaker diarization to label each speaker
- Remains in the meeting until the meeting ends or you cancel it from the dashboard
- Begins processing immediately when the meeting ends
What the bot does not do
- Does not have a camera - it is not visible on video
- Does not have a microphone - it cannot be unmuted and cannot speak
- Does not send chat messages of any kind
- Does not react to anything said in the meeting
- Does not capture video or screen shares
- Does not access files shared during the meeting
Visibility to your clients: The bot is visible in the participant list. Your clients can see it. This is by design - invisible recording is ethically and legally problematic. You should tell your client at the start of every call that it is being recorded by Audora. See the Recording consent section.
What happens if the bot loses connection mid-call
If the bot drops from the meeting due to a network issue, it attempts to rejoin automatically up to three times over a 5-minute window. If it cannot rejoin, it stops attempting and you are notified by email. The transcript and outputs cover only the portion of the call where the bot was present.
Partial call processing is treated the same as a full call - it counts against your monthly limit, and all outputs are generated from whatever transcript was captured.
What happens if the host removes the bot
If the meeting host removes the bot, it cannot rejoin - this is a hard platform restriction on Zoom and Teams. On Google Meet, the bot is removed silently and similarly cannot rejoin. Processing runs on whatever was captured before removal.
If you anticipate that your client might not want a recording bot on the call, discuss it before the call rather than discovering it mid-session.
Muting the bot
The bot has no microphone and cannot be muted in the traditional sense. If the host mutes "all participants," this has no effect on the bot's ability to listen - muting in video conferencing applies to microphones, not to audio capture by participants.
Platform-specific details
Zoom
The bot joins via any shareable Zoom meeting link. It works with:
- Password-protected meetings - you provide the password when creating the invite in the dashboard
- Waiting rooms - the bot waits and joins when the host admits it (up to 15-minute wait limit)
- Zoom Free and Zoom Pro accounts
- Recurring meetings - each session is treated as a separate call
Zoom waiting rooms: If your Zoom meeting has a waiting room enabled, you need to admit the "Audora" participant manually when starting the call. This takes about 3 seconds. If you forget to admit it, the bot times out after 15 minutes.
Google Meet
The bot joins via a Google Meet URL or via a calendar event that contains a Meet link. It works with:
- Personal Google accounts (gmail.com)
- Google Workspace accounts (any domain)
- Meetings where the host has enabled "Quick access" - no admission required
- Meetings where the host has disabled Quick access - the bot requests entry and waits for the host to admit it
Google Meet admission: If the meeting organizer has turned off Quick access, you will see an "Audora is waiting to join" notification from Google Meet. Click Admit to let it in. If you do not admit it, the bot times out after 15 minutes.
Microsoft Teams
The bot joins Teams meetings as an external participant. It is available on Studio and Agency plans. It works with:
- Teams Free accounts
- Teams with Microsoft 365 (any plan)
- Meetings where external guests are allowed - this is the default Teams setting
Teams external guest policy: Some organizations lock down Teams to prevent external participants from joining. If your client's organization has this restriction, the bot will be unable to join their hosted meetings. In this case, you will need to host the meeting yourself from your account, where the external guest policy is under your control, and invite your client.
| Feature | Zoom | Google Meet | Teams |
|---|---|---|---|
| Plans | All | All | Studio, Agency |
| Password-protected meetings | Yes | N/A | N/A |
| Waiting room support | Yes | Yes (Quick access) | Limited |
| Admin setup required | No | No | No |
| External participant restrictions | Rare | Rare | Possible |
Recording consent
This is a legal matter, not just a policy matter. Recording calls without proper consent is illegal in many jurisdictions. Audora cannot verify consent on your behalf. You are fully responsible for compliance.
Why this matters
Audora records audio from your video conference calls in order to generate transcripts. Recording laws vary significantly by country, state, and province. Violating these laws can result in civil liability or criminal charges.
One-party vs. all-party consent
The key distinction in recording law is between one-party and all-party (two-party) consent jurisdictions:
One-party consent
Only one person on the call needs to consent to the recording. Since you are consenting by using Audora, you may generally record without informing other participants. Most US states fall into this category.
All-party consent
Every participant must actively consent before recording begins. This includes California, Florida, Illinois, and most countries outside the US. Failure to obtain all-party consent is illegal regardless of your location if any participant is in one of these jurisdictions.
The safe default: Always treat every call as an all-party consent situation. Tell every participant at the start of every call that it is being recorded. This is legal, professional, and takes about 10 seconds.
Practical guidance
The most reliable way to obtain consent is verbal notice at the start of the call:
"Just to let you know, I'm recording this call with Audora to automatically generate our project spec. Is that okay with you?"
Most clients will say yes immediately. Some may ask what Audora is. Tell them it joins the call and generates the project documentation - transcripts, requirements, task list, and statement of work. This is typically well-received because it means they get a professional deliverable faster.
Audora's role
The Audora bot is visible in the participant list. Many video platforms also display a recording indicator when a bot joins. Neither of these substitutes for explicit consent - they are notifications, not consent mechanisms. You must obtain consent through your own communication with participants.
Audora's Terms of Service require you to comply with all applicable recording laws. Accounts where we have credible evidence of recording without required consent will be suspended.
Outputs overview
Every processed call produces six structured documents. All six are generated from the same transcript but serve different audiences and purposes. Understanding what each one is for helps you know when to edit, when to push, and when something looks wrong.
| Output | Audience | Purpose | Typical length |
|---|---|---|---|
| Transcript | Internal | Verbatim record of the call | Full call length |
| Client brief | Internal / client-facing | Plain-language summary of what the client wants | 300–600 words |
| Requirements doc | Internal / builder | Structured list of all functional requirements | 500–1,500 words |
| PRD | Build tools | Full product requirements document for AI coding tools | 800–2,500 words |
| Task breakdown | Internal / project tools | Sequenced tasks with time estimates | 10–40 items |
| SOW | Client-facing | Formal scope of work document | 400–900 words |
Working with outputs in the dashboard
All outputs are available in the call detail view in the dashboard. From here you can:
- Read - view the full document in the dashboard
- Edit - make changes directly in the dashboard before pushing or downloading. Edits are auto-saved.
- Download - download as a Markdown or plain text file
- Push - send to a connected integration with one click
- Copy - copy the full document to clipboard
Edit before you push. Audora generates from what was said in the call. Review each output before pushing to a client-facing tool or sending to a client. The SOW in particular should always be reviewed - clients sometimes make requests in calls that you should clarify or negotiate before committing to in a written document.
Transcript
What it is
The transcript is a full, word-for-word record of the call with each line attributed to the speaker who said it. It is the source of truth for all other outputs - everything Audora generates in the other five documents comes from information in the transcript.
Format
Transcripts are structured as speaker-labeled dialogue:
Shaban (00:01:14)
So the main thing I need is a dashboard where the logistics team can see
all active shipments in real time. Right now they're checking five different
spreadsheets and it's a nightmare.
Client (00:01:38)
Exactly. And it needs to work on mobile - most of them are in the warehouse,
not at a desk. So whatever you build, the mobile experience has to be solid.Each line includes the speaker name and the timestamp of when they started speaking.
Speaker identification
Audora assigns speaker labels based on voice diarization - the process of identifying which segments of audio belong to which person. On the first processed call with a new set of participants, speakers are labeled generically (Speaker 1, Speaker 2, etc.) until you assign names.
To assign names, open the transcript in the dashboard and click any speaker label. A field appears where you can type the person's name. Audora remembers this association for future calls where the same voice appears.
Improving speaker accuracy: Speaker identification accuracy improves when participants use quality microphones and are not talking over each other. Calls with more than four concurrent speakers or significant background noise have lower diarization accuracy.
Transcript search
The full text of all transcripts is searchable from the dashboard's search bar. Search returns matches across all calls, not just the current one. Useful for finding what a client said about a specific feature three weeks ago, or locating a particular project discussion.
Editing the transcript
You can edit the transcript directly in the dashboard. Edits are saved automatically. If you correct errors in the transcript, regenerating the outputs from an edited transcript produces more accurate results.
Regenerating outputs: After editing a transcript, click Regenerate outputs in the call detail view. This re-runs all six outputs from the corrected transcript. The previous outputs are replaced.
Client brief
What it is
The client brief is a concise, readable summary of the client's goals, expectations, and constraints as expressed in the call. It is written in plain English - not in technical spec language, and not as a summary of the conversation. It is a statement of what the client is trying to accomplish.
What it contains
- Client context - who they are, what they do, and why they need this built
- The core problem - what situation they are currently in that makes this project necessary
- What they want - the product or feature they asked for, in their own framing
- Key constraints - deadline pressure, budget signals, platform requirements, non-negotiables mentioned in the call
- What success looks like - how the client described a good outcome
When to use it
The client brief is most useful for:
- Sharing with team members who were not on the call and need to understand the client situation quickly
- Referencing before a follow-up call to re-orient yourself on the client's perspective
- Sending to a client as a lightweight confirmation of your understanding - "here's what I took away from our call"
Sending the brief to the client: The client brief is written to be shareable. If you send it to your client and ask them to confirm it captures what they need, you accomplish two things at once - you demonstrate professionalism, and you catch any misunderstandings before they become scope problems.
Requirements document
What it is
The requirements document translates what the client said - often in natural, imprecise language - into discrete, numbered functional requirements. Each requirement is a specific, testable statement of what the product must do.
Structure
Requirements are grouped by feature area and numbered sequentially. Audora also flags requirements that are ambiguous or potentially problematic:
Dashboard - Core
REQ-001: The system shall display all active shipments in a single view
REQ-002: Shipment status must update in real time (no manual refresh required)
REQ-003: Each shipment row shall show: ID, origin, destination, carrier, status, ETA
Mobile Experience
REQ-004: The dashboard must be fully functional on mobile browsers
REQ-005: Touch targets shall be appropriately sized for warehouse use (gloves)
⚠ Ambiguous - REQ-006: "some kind of alert when something goes wrong"
Client did not specify: what triggers an alert, how alerts are delivered,
or who receives them. Clarification needed before estimating.Ambiguity flags
Requirements marked with ⚠ are cases where the client expressed a need but did not specify it precisely enough to build from without further clarification. These are not errors - they are signals that something needs a follow-up question before the build starts. Resolving them before writing the PRD produces significantly better outcomes.
Exclusions
Audora also notes explicit exclusions - things the client said are out of scope or not needed in the first version. These are included in the requirements document as a reference to prevent scope creep.
PRD (Product Requirements Document)
What it is
The PRD is the most comprehensive output. It is designed to be the opening context for an AI coding session - everything a coding AI needs to understand the product it is about to build, without additional prompting.
What it contains
- Problem statement - why this product is being built and for whom
- User types - who uses the product and what their relationship to it is
- Core features - each major feature with description and acceptance criteria
- Technical constraints - platform requirements, integration requirements, performance requirements mentioned in the call
- Explicit exclusions - what is not being built in this version
- Success criteria - how to know when the product is done
How it is used with build tools
The PRD is pushed to different tools in different formats:
| Tool | How the PRD is delivered |
|---|---|
| Cursor | Written to .audora/spec.md in your repository. Open Cursor on the repo and it loads as context. |
| Lovable | Formatted as an optimized Lovable prompt. A pre-loaded project link is generated and available in the dashboard. |
| Bolt | Same mechanism as Lovable - a pre-loaded Bolt project link generated from the PRD. |
Editing the PRD before pushing
The PRD is the output you are most likely to want to edit before pushing. Common edits include:
- Adding technical decisions you have already made that were not mentioned in the call (tech stack, hosting, etc.)
- Resolving ambiguous requirements flagged in the requirements document
- Removing requirements you have decided to defer to a later version
- Adding context about the codebase if you are extending an existing project
The PRD is a starting point. Audora documents what was discussed. You know the technical context. The best PRDs are ones where Audora captures the client requirements and you add the technical decisions. Together, they give the coding AI a complete picture.
Task breakdown
What it is
The task breakdown is a sequenced list of every development task required to ship the project. Tasks are ordered in build sequence - dependencies come first. Each task includes a title, a brief description, and a time estimate in hours.
Task structure
1. Project setup and environment configuration (2h)
Initialize repository, configure build tooling, set up deployment pipeline.
2. Database schema design (3h)
Design shipment, user, and alert tables. Write migrations.
3. Shipment data model and API (4h)
REST endpoints for shipment CRUD. Real-time update architecture decision.
4. Dashboard UI - shipment list view (6h)
Table component with sorting, filtering, and status indicators.
5. Real-time update integration (4h)
WebSocket or polling implementation. Connect to shipment API.
...Time estimates
Time estimates are generated from the complexity of each task as described in the call and inferred from the requirements. They are starting points, not commitments. For projects involving technologies or integrations that were not discussed in the call, estimates may be off. Treat them as a relative complexity guide rather than a precise schedule.
How tasks are pushed to integrations
| Integration | What gets created |
|---|---|
| Linear | One issue per task. Issue title from task name, description from task description. Assigned to backlog. Priority: medium by default. |
| Notion | A checklist database with one row per task. Columns: task name, description, estimate, status (default: To Do). |
Statement of Work
What it is
The Statement of Work is a formal document defining what you are agreeing to deliver. It is written in professional language and structured as a business document - not a technical spec. It is designed to be sent directly to a client as a formal record of scope.
What it contains
- Project overview - what is being built and for whom
- Scope of work - what is specifically included in this engagement
- Deliverables - what you will hand over at completion
- Timeline - estimated duration based on the task breakdown
- Out of scope - explicit list of what is not included
- Assumptions - things you are assuming to be true; if they change, scope may change
- Revision terms - how revisions and change requests are handled
Always review the SOW before sending it to a client. Audora writes what the client asked for. That is not always the same as what you should formally commit to. Clients say things in discovery calls - "can you also add X?" - that should be scoped separately rather than included in the initial SOW. Review every SOW with the lens of: is this what I am actually willing to deliver for this price?
Common edits to the SOW
- Adding your pricing or payment terms (Audora does not add prices - you know your rates)
- Removing features the client mentioned casually that you want to treat as separate scope
- Tightening the revision terms if you want more structure around change requests
- Adding your standard legal boilerplate if you have it
Output quality guide
What determines output quality
Audora generates from what was said in the call. The quality of the outputs is a direct function of the quality of the conversation. The best outputs come from calls where:
- The client clearly articulates specific requirements - "it needs to do X" rather than "I want something like Y but better"
- Constraints are stated explicitly - platform, timeline, integration requirements
- Out-of-scope items are discussed - what is not being built in this version
- Follow-up questions get specific answers rather than vague confirmations
Structuring calls for better outputs
You do not need to change how you run calls. But if you want consistently excellent outputs, these patterns help:
- Ask for specifics on each feature - "what does that look like in practice?" gets you usable requirements. "Does that make sense?" gets you nothing.
- Name the constraint explicitly - if it needs to be mobile, say "mobile" not "accessible from the warehouse." Audora picks up explicit statements better than inferences.
- State what is not in scope - say out loud "we're not doing X in this version." Audora captures this and includes it in the SOW and requirements as an explicit exclusion.
- Confirm timeline expectations - if the client has a deadline, say it. Even a rough one ("before the end of the quarter") gives Audora something to work with in the SOW.
When outputs look wrong
If an output contains something inaccurate or missing:
Check the transcript first
If the transcript is wrong, the outputs will be wrong. Transcription errors are the most common cause of bad outputs - a misheard word in a key requirement can cascade into an incorrect PRD.
Correct the transcript and regenerate
Edit the transcript in the dashboard, then click Regenerate outputs. All six outputs are regenerated from the corrected transcript.
Edit the output directly
For minor issues - a poorly phrased sentence, a missing detail you want to add - edit the output directly in the dashboard. No need to regenerate everything for small corrections.
Integrations overview
How integrations work
Integrations connect Audora to your build and project management tools. Once connected, Audora pushes outputs automatically after every call - or you can push manually from the call detail view per output, per integration.
All integrations are configured in Settings → Integrations. Each integration shows its connection status, the last time it was used, and a test push button.
Push behavior
By default, all connected integrations receive a push after every processed call. You can change this behavior:
- Auto-push all integrations - default. Every call pushes to every connected integration.
- Manual push only - disable auto-push in Settings → Integrations. You trigger each push manually from the call detail view.
- Per-call override - from the call detail view, you can enable or disable any integration for that specific push before pushing.
Integration authorization
Most integrations use OAuth - you click Connect, authorize Audora in the third-party application, and are redirected back. Audora stores a refresh token to maintain the connection. If you revoke access from the third-party application's side, Audora loses the connection and will notify you the next time a push fails.
Cursor integration
How it works
The Cursor integration uses your GitHub account to write the PRD and task breakdown to a repository of your choice. When you open Cursor on that repository, the spec is immediately available as project context - no manual loading, no copy-pasting.
Connecting
Go to Settings → Integrations → Cursor
Click Connect. You are redirected to GitHub OAuth.
Authorize Audora on GitHub
Audora requests permission to write to repositories. You can grant access to all repositories or select specific ones. Granting to specific repositories is recommended.
Select your default repository
Choose which repository Audora should write to by default. You can change this per call in the push settings.
What gets written to the repository
Audora creates an .audora/ directory in the root of your repository and writes two files:
| File | Contents |
|---|---|
.audora/spec.md | The full PRD. Cursor uses this as context when you open the project. |
.audora/tasks.md | The task breakdown. Formatted as a Markdown checklist. |
Add .audora/ to your .gitignore if you do not want spec files committed to your repository history. Audora writes to the files regardless - adding it to .gitignore only affects whether the files appear in your commits.
Pushing to a different repository per call
From the call detail view, before pushing, click the settings icon next to Cursor. Select a different repository from your connected GitHub account. This overrides the default for this push only.
Lovable integration
How it works
The Lovable integration requires no OAuth connection. After each call, Audora formats the PRD as an optimized Lovable prompt and generates a shareable project link. Clicking the link from the dashboard opens Lovable with the full spec pre-loaded as the initial prompt - you land directly in a new project with your requirements already there.
No connection required
Because Lovable's project links are self-contained, Audora does not need access to your Lovable account. The link generates from your PRD and opens in Lovable in your browser. Your Lovable account and projects are not visible to Audora.
Using the generated link
Open the call detail view
After a call is processed, open it from the dashboard.
Click the Lovable push button
Audora generates the Lovable link and opens it in a new browser tab. You are taken directly to a new Lovable project with the spec as the opening prompt.
Review and start building
Check that the prompt looks right, then start Lovable. You can edit the prompt before submitting if you want to add or remove anything.
Edit the PRD before pushing. Since the Lovable prompt is generated from your PRD, editing the PRD in the dashboard before clicking push gives you a more tailored Lovable prompt. Add technical constraints, remove requirements you want to defer, or add context about your preferred stack.
Bolt integration
How it works
The Bolt integration works identically to the Lovable integration. No connection is required. Audora formats the PRD for Bolt's prompt structure and generates a pre-loaded project link. Clicking it opens a new Bolt project with your spec already loaded.
Lovable vs Bolt - which to use: Both accept the same underlying spec from Audora. The choice comes down to the project type and your preference. Bolt tends to work better for full-stack applications with complex backend logic. Lovable tends to work better for UI-heavy applications with clean data models. Audora generates a link for both - you can try both and see which produces a better starting point for a given project.
Linear integration
StudioAgency
How it works
The Linear integration creates a new project in your Linear workspace and populates it with issues drawn from the task breakdown. Each task becomes a Linear issue with a title, description, and default settings. Your project board is ready without any manual issue creation.
Connecting
Go to Settings → Integrations → Linear
Click Connect. You are redirected to Linear's OAuth flow.
Authorize Audora in Linear
Grant Audora permission to create projects and issues in your workspace. Audora requests write access to projects and issues only - it cannot read your existing issues or modify your workspace settings.
Select your default team
Choose which Linear team new projects should be added to. You can override this per call.
What gets created in Linear
| Item | Details |
|---|---|
| Project | Named after the client and call date (e.g., "Acme Corp - March 24"). Created in your selected team. |
| Issues | One issue per task. Title from task name, description from task description. Assigned to backlog. Priority: medium. |
Customizing Linear push settings
From Settings → Integrations → Linear, you can configure:
- Default team for new projects
- Default assignee for new issues (or leave unassigned)
- Default issue state (Backlog or To Do)
Notion integration
StudioAgency
How it works
The Notion integration creates a new page in a Notion database of your choice. The page contains the complete project package - PRD, requirements document, task checklist, and SOW - organized as nested blocks. The page is shareable with a Notion link.
Connecting
Go to Settings → Integrations → Notion
Click Connect. You are redirected to Notion's OAuth flow.
Authorize Audora in Notion
Select the pages and databases Audora can access. Grant access to the database where you want project pages to be created. Audora can only access pages you explicitly share with it.
Select your default database
Choose which Notion database new project pages should be added to. The database should have at minimum a Name property and a plain text or rich text body area.
What gets created in Notion
A new page is created in your selected database with the following structure:
- Page title - client name and call date
- PRD section - full PRD as rich text blocks
- Requirements section - numbered list of all requirements
- Task checklist - each task as a Notion to-do block with checkbox
- SOW section - full SOW as rich text blocks
Plans and pricing
| Plan | Monthly | Annual | Calls/mo | Seats |
|---|---|---|---|---|
| Solo | $49 | $44 | 15 | 1 |
| Studio | $99 | $89 | 40 | 3 |
| Agency | $199 | $179 | 100 | 8 |
What every plan includes
- All six outputs per call (transcript, brief, requirements, PRD, task breakdown, SOW)
- Zoom and Google Meet bot
- Cursor, Lovable, and Bolt integrations
- Full dashboard access
- Email support
Studio and Agency additionally include
- Microsoft Teams bot
- Linear integration
- Notion integration
- Team seats (3 for Studio, 8 for Agency)
- Priority support (Studio), dedicated Slack support (Agency)
What counts as a call
A call is counted when the Audora bot successfully joins a meeting and produces at least a partial transcript. Calls where the bot joined a waiting room but was never admitted and timed out do not count. Calls where processing fails due to a technical error on our end do not count and are flagged in the dashboard.
Call limits and overages
When you reach 80% of your monthly call limit, you receive an email notification. When you hit the limit, the bot stops joining new calls until your billing cycle resets or you upgrade. There are no overage charges - you simply cannot use more calls than your plan allows.
Managing your billing
Upgrading your plan
Go to Settings → Billing → Change plan. Select the new plan and confirm. Upgrades take effect immediately - you gain access to the new plan's features and call limit right away. You are charged the prorated difference for the remainder of your current billing cycle, then billed at the new rate from the next cycle.
Downgrading your plan
Go to Settings → Billing → Change plan. Select a lower plan and confirm. Downgrades take effect at the start of your next billing cycle. You retain your current plan's features and call limit until then. Unused calls from your current cycle do not carry over.
Downgrading from Studio or Agency to Solo removes access to the Teams bot, Linear, and Notion integrations on the next billing cycle. Any existing Linear projects or Notion pages created by Audora are not deleted - they remain in your Linear and Notion accounts. You simply cannot push new ones.
Switching between monthly and annual billing
You can switch from monthly to annual billing at any time. The switch takes effect at the start of your next billing cycle. You cannot switch from annual to monthly before the annual period ends - contact support if you need an exception.
Updating your payment method
Go to Settings → Billing → Payment method. Add a new card or update the existing one. The new payment method is used for all future charges immediately. Previous invoices are not affected.
Invoices
All invoices are available in Settings → Billing → Invoice history. Each invoice can be downloaded as a PDF. Invoices are also emailed to your account email address after each payment.
Failed payments
If a payment fails, you receive an email notification and we retry the charge automatically after 3 days, then again after 7 days. If the third attempt fails, your account is moved to a restricted state - you can log in and view your data, but the bot will not join new calls and integrations will not push. Resolve the payment issue from Settings → Billing to restore full access.
Free trial
What you get
All new accounts start with a 14-day free trial on the Studio plan. This means you have access to everything on Studio - 40 calls, 3 seats, all integrations including Teams, Linear, and Notion - for 14 days with no payment required.
When the trial ends
If you have not added a payment method by the end of day 14, your account moves to a read-only state:
- You can still log in and view all your calls and outputs from the trial
- The bot will not join new calls
- Integrations will not push
- Your data is retained for 30 days, then permanently deleted
To continue using Audora, go to Settings → Billing → Add payment method and select a plan. Access is restored immediately after payment.
Run Audora on a real client call within the first 48 hours. Users who try it on a real call in the first two days are much more likely to find it immediately useful. The outputs from a real client conversation - compared to a test - make the value obvious.
Team seats
StudioAgency
Seat limits
Studio plans include 3 seats. Agency plans include 8 seats. Each seat is one user account. All seats share the workspace's monthly call limit - if you have 3 seats and your plan includes 40 calls, the total across all three users is 40 calls per month.
Inviting a team member
Go to Settings → Team → Invite member
Enter the email address of the person you want to invite.
Select their role
Choose Owner or Member. See the Permissions section for details on what each role can do.
Send the invite
An invitation email is sent. The invite link expires after 7 days. If it expires before they accept, resend it from Settings → Team.
Removing a team member
Go to Settings → Team, find the member, and click Remove. Their access is revoked immediately - they are signed out of any active sessions. Their past call outputs remain in the workspace and are accessible to the account owner. Removing a member frees up a seat that can be filled with a new invite.
Call visibility between team members
By default, team members can only see calls and outputs from sessions where the bot was invited under their own account. The workspace owner can see all calls across the team from the dashboard. This default cannot be changed in the current version.
Permissions
| Action | Owner | Member |
|---|---|---|
| Invite the bot to calls | Yes | Yes |
| View own call outputs | Yes | Yes |
| View all team member calls | Yes | No |
| Edit outputs | Yes | Yes (own calls only) |
| Push to integrations | Yes | Yes (own calls only) |
| Manage integrations | Yes | No |
| Invite team members | Yes | No |
| Remove team members | Yes | No |
| Manage billing | Yes | No |
| View invoices | Yes | No |
| Delete the account | Yes | No |
Multiple owners: A workspace can have multiple owners. If you want a team member to have full admin access, set their role to Owner when inviting them, or change their role from the team settings after they have joined.
How your data is handled
What we collect
- Call audio - captured by the bot during your meeting
- Transcripts - generated from the audio
- Generated outputs - the six documents produced per call
- Account information - name, email, password (hashed), billing details
- Usage data - which features you use, call counts, processing times
- Calendar data - meeting times and conference links, if you connect Google Calendar
Where it is stored
All data is stored on servers in the United States. Data is encrypted at rest (AES-256) and in transit (TLS 1.3). Each account's data is stored in an isolated environment - there is no shared storage between accounts.
Who can access it
Your data is accessible to:
- You (and your team members, subject to permissions)
- Audora staff, only when required for technical support you have requested, and only through audited, time-limited sessions
- Our infrastructure subprocessors, under contractual data processing agreements that restrict use to providing the service only
No one else. We do not sell, share, or trade your data.
Retention and deletion
| Data type | Retention period |
|---|---|
| Call audio recordings | Deleted within 72 hours of transcription completing |
| Transcripts | Retained for subscription duration + 30 days after account closure |
| Generated outputs | Retained for subscription duration + 30 days after account closure |
| Account information | Retained for subscription duration + 30 days after account closure |
| Billing records | 7 years (legal requirement) |
| Access logs | 12 months |
What happens when you cancel
When you cancel your subscription, your account enters a 30-day grace period. During this period you can log in and export any data you want to keep. After 30 days, all data except billing records is permanently and irreversibly deleted.
Requesting early deletion
If you want your data deleted before the 30-day period ends, email support@audora.agency with "Data deletion request" in the subject line. We will process the deletion within 72 hours and confirm by email when it is done.
AI and your data
What AI does in Audora
Audora uses large language models for two things: transcription (converting audio to text) and document generation (converting the transcript into the six structured outputs). The AI runs in our infrastructure and is not directly exposed to users.
Your data is not used to train AI models
We do not use your call recordings, transcripts, or generated outputs to train, fine-tune, or evaluate any AI model - ours or any third party's. This is a contractual commitment with our AI infrastructure providers, not just a policy preference on our part.
This means: nothing you say in a client call, nothing in your transcripts, and nothing in your generated PRDs or SOWs will ever appear in an AI training dataset or improve an AI model that other users interact with.
How AI processing works
When Audora processes a call, the transcript is sent to an AI model in an isolated, stateless context. The model sees only the transcript for that call - it has no memory of previous calls from your account or any other account. After generating the outputs, the processing context is discarded. The model retains nothing.
Third-party AI providers
Our AI infrastructure providers are under data processing agreements that explicitly prohibit them from using your data for any purpose beyond providing the transcription and generation services to Audora. These agreements are auditable. If you need copies of the relevant clauses for your own compliance review, contact support.
Bot issues
The bot did not join my call
Check the dashboard for a bot status notification - it shows the reason it did not join. Common causes:
- Meeting link was incorrect or expired - Zoom links with password parameters sometimes get truncated when pasted. Copy the full link including the password parameter.
- Waiting room timeout - the bot waited 15 minutes without being admitted. Check your meeting settings and admit the bot as soon as it appears in the waiting room.
- Meeting had not started - if you invite the bot to join at a specific time and the meeting is still in the waiting room state (host has not started it), the bot will wait up to 15 minutes before timing out.
- External participants blocked (Teams) - your client's organization has disabled external participants. You need to host the meeting from your own account.
The bot joined but was removed immediately
Someone on the call - typically the host - removed the bot. This usually happens when the client was not expecting it. Tell clients in advance that Audora will join the call. A brief message before the call ("I use Audora to auto-generate the project spec - it'll join as a participant to record") prevents most removals.
The bot joined but I see no outputs
If the call is still processing, wait a few minutes. For calls under 90 minutes, processing should complete within 2–3 minutes of the call ending. If it has been more than 10 minutes with no output and no processing error notification, email support@audora.agency with the call date, time, and platform.
The bot joined late and missed the start of the call
If the bot joined after the call started, the transcript begins from when it joined. The outputs cover only that portion of the call. For important calls, schedule the bot to join 2–3 minutes before the meeting start time using the scheduled join option in the dashboard.
Output issues
The transcript has errors
Transcription errors are the most common cause of output problems. They happen most often with: heavy accents, technical jargon specific to a narrow domain, names of people or companies (especially unusual ones), and calls with background noise or poor microphone quality. Edit the transcript directly in the dashboard, then click Regenerate outputs to produce corrected versions of all six documents.
The PRD is missing features the client mentioned
First check the transcript - if the feature was transcribed correctly, check the requirements document. If it appears in the requirements but not the PRD, click Regenerate outputs. If it is not in the transcript at all, the client said it during a period where audio quality was poor or the bot had dropped. You will need to add it manually to the PRD in the dashboard.
The SOW includes something I do not want to commit to
Edit the SOW directly in the dashboard before downloading or sending it to the client. The SOW is a starting point - Audora writes what was discussed, not what you have agreed to. Review it the same way you would review any contract before signing.
The task estimates seem wrong
Task estimates are generated from the complexity implied by the call discussion. They are starting points for planning, not firm commitments. If a task involves a technology or integration that was not discussed in the call, Audora cannot estimate it accurately. Edit estimates directly in the dashboard before pushing to Linear or Notion.
All six outputs look empty or generic
This typically happens when the transcript is very short - either the call was very brief, the client did not articulate specific requirements, or the bot joined late and only captured a few minutes of audio. It can also happen if the call was mostly social conversation rather than requirements discussion. In this case, there is not enough source material to generate detailed outputs - that is not a bug.
Integration issues
Push failed - what does that mean?
A push failure means Audora attempted to send an output to a connected integration and the integration rejected it. Common causes:
- Authorization expired - the OAuth token Audora uses to access the integration has expired or been revoked. Go to Settings → Integrations, find the affected integration, and click Reconnect.
- Integration permissions changed - if you changed permissions in Linear, Notion, or GitHub after connecting, Audora may no longer have the access it needs. Reconnect the integration to re-authorize with current permissions.
- Target not found - the repository, database, or team you configured no longer exists or was renamed. Update the integration settings.
The Cursor push wrote to the wrong repository
From the call detail view, find the Cursor push and click Push again. Before confirming, click the settings icon to select a different repository. You can also update your default repository in Settings → Integrations → Cursor.
Linear issues are being created in the wrong team
Your default Linear team may not be the one you wanted for this project. From the call detail view, before pushing, click the settings icon next to Linear and select the correct team. Update your default in Settings → Integrations → Linear to change it globally.
The Notion page was created but I cannot find it
Check which Notion database Audora is configured to use in Settings → Integrations → Notion. Open that database in Notion - the page should be there. If your Notion database has filters applied that are hiding new entries, check the filter settings.
Billing issues
I was charged but I cancelled during the trial
The 14-day trial requires no payment method. If you added a payment method and selected a plan during the trial, you were charged at the end of the trial period or immediately if you selected a plan before the trial ended. If you believe you were charged in error, email support@audora.agency within 7 days of the charge.
I want a refund
Refunds are available within 7 days of your first payment, provided you have used less than 20% of your plan's call limit during that period. Email support@audora.agency with "Refund request" in the subject line, your account email, and the date of the charge. We respond within 3 business days. See the full Refund Policy.
My payment failed and my account is restricted
Go to Settings → Billing → Payment method. Update your card or add a new one. After a successful payment, full access is restored immediately. We retry failed payments automatically (after 3 days and after 7 days) - if the payment succeeds on a retry, your access is restored automatically.
I do not recognize a charge
All Audora charges appear as "Audora" on your statement. Email support@audora.agency with the charge amount and date. We will identify the invoice and confirm it matches your account activity.
Contact support
Email support
Email support@audora.agency. We respond within one business day on Solo and Studio plans. Agency plan users have access to a dedicated Slack channel where response times are faster.
What to include in your message
The more context you give us, the faster we can help. For most issues, include:
- Your account email address
- The date and approximate time of the call (if the issue is related to a specific call)
- The meeting platform (Zoom, Meet, or Teams)
- What you expected to happen and what happened instead
- Any error messages shown in the dashboard
Response times
| Plan | Support channel | Response time |
|---|---|---|
| Solo | Within 1 business day | |
| Studio | Priority email | Within 1 business day |
| Agency | Dedicated Slack | Within a few hours during business hours |