AnythingLLM Meeting Assistant: The Complete Guide

AnythingLLM's Meeting Assistant automatically transcribes, summarizes, and indexes your video calls so you never lose important information. It's free, runs on your own hardware with no cloud dependency, and turns meeting recordings into searchable knowledge. Here's exactly how it works and why it matters if you're tired of paying Otter or Fireflies.

What Is AnythingLLM's Meeting Assistant?

AnythingLLM's Meeting Assistant captures live meetings and recordings, converts audio to text, identifies speakers, and generates summaries without uploading anything to the cloud. Built into the desktop app, it's a single feature among many in AnythingLLM, not a standalone product. That simplicity is the point: your meeting data stays local, accessible only to you.

The tool solves three concrete problems. First, it eliminates expensive per-minute billing from SaaS meeting assistants like Otter ($600+ annually) or Fireflies ($120+ annually). Second, it keeps your meeting audio private. No cloud processing means no logs, no data retention policies, no compliance audits. Third, it turns transcripts into raw material for your AI knowledge base: ask your LLM about what was actually said in meetings, not summaries someone else wrote.

You need to decide whether this matters to your workflow. If you run a 10-person team with 50 meetings a week, the cost savings alone justify switching. If your calls involve contracts, customer details, or proprietary strategy, privacy might be non-negotiable. AnythingLLM does both at once.

Key Features: What It Can Do

The Meeting Assistant ships with four core capabilities: recording, transcription, speaker identification, and summarization. Each one plugs directly into AnythingLLM's broader system so your meetings become part of your RAG context.

Recording and Real-Time Transcription

You start recording from within the AnythingLLM app or join a meeting platform (Zoom, Google Meet, Teams) directly and hit record. Audio transcribes in real time, so you see text appearing as people speak. Processing speed is measured in single-digit minutes: a three-hour meeting typically takes two to five minutes to fully transcribe, depending on hardware. That's faster than most of your attendees can transcribe a meeting manually, and infinitely faster than hiring a transcriptionist.

Speaker Identification with Diarization

Diarization is the technical term for assigning speakers to segments of audio. AnythingLLM offers three modes: no diarization (just one transcript block), basic diarization (identifies speaker changes but not who is who), and full diarization (attempts to tell Speaker 1 from Speaker 2 automatically). Full mode requires more compute but gives you clean, attributed transcripts: "Speaker A: Let's ship this Tuesday. Speaker B: I need Thursday for testing."

You can manually edit speaker labels afterward, which is important because automatic diarization often gets names wrong on low-quality audio or when multiple people talk simultaneously. The interface lets you rename "Speaker 1" to "Alice" before saving.

Meeting Summaries with Custom Templates

Summaries are optional and on-demand. AnythingLLM provides three built-in templates: general, sales call, and engineering call. Each one generates a prompt that feeds the full transcript to your local LLM and extracts key points, action items, decisions, and next steps. You can create your own templates for board meetings, 1:1s, customer interviews, or any call type that benefits from a standard structure.

Templates matter because a generic summary of a technical design review tells you nothing, but a summary asking for "decisions made, open questions, assigned owners, and blocking issues" is immediately actionable. You can regenerate summaries with different templates after the fact.

Search and Follow-Up

Transcripts are indexed so you can search them like documents. This is where the integration with your knowledge base becomes powerful. If you ask your AnythingLLM chatbot "What did we commit to on the vendor contract in last month's legal call?", it can search across all meeting transcripts and pull the exact quote.

Action items can be extracted and viewed separately, and AnythingLLM has an "agentic" follow-up mode that can automatically create Slack notifications, Jira tickets, or other follow-ups based on meeting content. This requires additional configuration but removes the "oh, I forgot that was supposed to be done" problem.

How to Set Up and Record Your First Meeting

Setup takes five minutes. Install AnythingLLM Desktop (free, open source), launch it, and navigate to the Meeting Assistant tab. Check which recording mode you want: system audio (captures everything your speakers output) or microphone (just you).

Step 1: Enable Meeting Capture

Go to Settings > Meeting Assistant and toggle the feature on. Choose your diarization mode. If you're unsure, start with basic diarization; you can upgrade to full mode and re-process later if needed. Select an LLM model to use for summaries; if you have a local model running (Mistral, Llama 2, etc.), pick it; otherwise pick the fastest available.

Step 2: Record a Meeting

For a Zoom or Google Meet call, join the call normally through your browser or app. Once the meeting starts, click the recording button in AnythingLLM. For a recording file you already have (MP4, MP3), import it directly and AnythingLLM will process it the same way as a live recording.

Step 3: Wait for Processing

Transcription runs in the background. Most meetings process while you're still taking notes. You'll see a progress bar: transcription, then speaker identification, then summary generation. Once complete, click into the meeting and review the transcript. Correct any speaker name mismatches or text errors.

Step 4: Create or Apply a Summary Template

Choose an existing template or write a custom prompt. Hit "Regenerate Summary" and let your LLM write it. The summary appears below the transcript. You can save it, export it as markdown, or share the link with teammates.

Speaker Identification: Which Diarization Mode Is Right for You?

Diarization is the feature that determines whether your transcript reads as a coherent dialogue or a wall of unattributed text. Understanding the three modes helps you pick the right tradeoff.

No Diarization: Speed Over Structure

No diarization means the entire meeting becomes one continuous block of text without speaker labels. It's the fastest option and uses minimal CPU. Use this if you're archiving meetings for compliance (you have audio for proof) or if you're only searching transcripts for keywords, not reading them for context. Downside: "Expand the timeline and increase the budget" is useless without knowing who said it.

Basic Diarization: Good Enough for Most Calls

Basic mode detects speaker changes and groups text by speaker, but doesn't attempt to name them. You see "Speaker 1: Expand the timeline. Speaker 2: Increase the budget." You manually edit the names afterward, which takes 30 seconds for a three-person call. It's the sweet spot: meaningfully structured without heavy compute cost. Most users here.

Full Diarization: Maximum Detail at Higher CPU Cost

Full mode tries to identify who is who automatically, using voice characteristics and model training. On clean audio with distinct voices, it works well. On noisy Zoom calls with multiple women or men, it often gets them swapped. Full mode also takes longer to process. Use it when automatic accuracy matters: legal depositions, recorded interviews with identifiable voices. Otherwise basic mode plus manual fixes is faster.

If diarization fails or produces garbled output, it's almost always an audio quality issue: background noise, music, crying babies, multiple people talking at once. Zoom's default audio compression makes this worse. If you can, use a USB headset for better source quality.

Customizing Summaries and Templates

Templates are the secret to getting useful summaries. A generic summary of an engineering standup produces platitudes. A standup summary asking specifically for "blockers, completed work, and priorities for tomorrow" produces a document your team can actually read.

Built-in templates are starting points. You can duplicate and modify any template or create one from scratch. The prompt is plain English describing what you want extracted. An example engineering template might read:

"Extract: (1) Completed work since last standup. (2) Current blockers preventing progress. (3) Priorities for the next 24 hours. (4) Any architecture or design decisions discussed. Return as markdown with clear sections."

You can make templates as specific as your call types. A customer support team might create a template asking for complaint theme, root cause, workaround applied, and escalation flag. A sales team might ask for deal name, stage, next step, and risk flag.

After writing a template, apply it to past meetings and review the output. If summaries are missing important details, refine the prompt. Templates are saved locally, so you build a library over weeks. After two months, you have templates for every call type your company runs.

How It Compares to Paid Alternatives

The meeting assistant space is dominated by SaaS tools charging per month or per minute. AnythingLLM is the open-source alternative. Here's how it actually stacks up.

Otter (Enterprise)

Otter charges $120-600 annually for unlimited meetings and real-time transcription. It's widely used and integrates with Zoom, Teams, Google Meet. It provides summaries, speaker identification, and AI highlights. Otter's main advantage is brand recognition and Slack/Salesforce integrations. Its disadvantage is that your audio is processed on Otter's servers, subject to their privacy policy and data retention terms. You're also paying for cloud storage and Otter's infrastructure.

AnythingLLM costs zero dollars. Meetings stay on your machine. There's no monthly bill to defend to finance, and no data leaving your network.

Fireflies (Mid-Market)

Fireflies charges $10-50 per user monthly and processes meetings on their cloud. It offers meeting search, integrations with Hubspot and Salesforce, and AI-generated action items. It's designed for teams. The main tradeoff: your meeting data is Fireflies' data, they can use it for model training (read their terms), and you're locked into their platform for search and integrations.

AnythingLLM stores nothing in the cloud and moves to a different platform instantly. Your data is portable.

Fathom (Chrome Extension)

Fathom records Zoom and Google Meet automatically via a browser plugin and charges nothing for transcription. That makes it cheaper than paid tools. But Fathom is closed source, the plugin runs code in your browser, and you don't control where recordings go. Read their privacy policy if you use it.

AnythingLLM is open source and auditable. You can review the code, compile it yourself, and run it in an air-gapped environment if needed.

Grain (Team Repository)

Grain ($20-100/month per team) focuses on building a searchable meeting library with integrations to Slack and other tools. It's built for larger teams. Meetings are stored on Grain's infrastructure.

AnythingLLM doesn't try to be a team repository; it's a personal tool. If you need team-wide search and Slack notifications, Grain is more mature. If you need privacy and cost control, AnythingLLM wins.

The Honest Comparison

If you're in a 50-person company running 100 meetings a week, Otter or Fireflies might make sense for the integrations and ease of setup. You pay per user or per minute but don't manage any infrastructure.

If you're a founder, solo consultant, or small ops team running 5-10 meetings daily, AnythingLLM is free and solves the problem completely. No contracts, no data leaving your machine, no monthly billing.

Mid-sized companies (10-50 people) sit in the middle. A team instance of AnythingLLM on managed infrastructure gives you the same privacy and cost control as self-hosted, but with someone else maintaining the servers.

Self-Hosted Deployment: Why It Matters

AnythingLLM Desktop runs on your laptop, but you can also deploy the Meeting Assistant to a shared server. This is useful if you're a 20-person team and want shared access to meeting transcripts without each person buying a machine powerful enough to run the LLM locally.

Deploying to a self-hosted server (Docker, Kubernetes, or bare metal) gives you three advantages over cloud SaaS alternatives.

First, data stays inside your network. Audio, transcripts, and summaries never leave your data center. If your industry requires this (healthcare, legal, finance, government), self-hosting is mandatory. Cloud SaaS tools promise this but funnel data through their infrastructure regardless.

Second, you control the LLM. You pick which model runs summaries: tiny Phi (2B parameters, fast, cheaper), medium Mistral (7B, balanced), or large Llama 2 (70B, slower but more accurate). Each gives different summary quality. Cloud tools pick the model for you.

Third, there's no per-meeting billing. Otter and Fireflies charge by the minute. Run 1,000 hours of meetings a year, and you're paying thousands. Self-host, and your cost is the server hardware and electricity. At scale, this is 80% cheaper.

Onsite deployment of AnythingLLM requires a machine with 8GB+ RAM and an optional GPU (Nvidia RTX or similar). Opsily offers managed AnythingLLM hosting that handles infrastructure, updates, and backups for you. This splits the difference: privacy and control without running your own servers.

Integrating the Meeting Assistant into your knowledge base workflow means recorded meetings feed into RAG queries. Ask your chatbot "What's our pricing strategy for enterprise?", and it searches across meeting transcripts, documents, and wikis. This only works if you own the data, which self-hosting guarantees.

Frequently Asked Questions

What audio formats and meeting platforms are supported?

AnythingLLM accepts MP3, MP4, WAV, and WebM files. For live recording, it hooks into Zoom, Google Meet, Microsoft Teams, and Slack calls. You can also record a meeting externally and import the file later. The app will process whatever audio format your system can play.

What are the hardware requirements?

Minimum: 8GB RAM, dual-core processor. Recommended: 16GB RAM, SSD storage, and an Nvidia GPU if you want summaries in under five minutes. On a MacBook Pro (M1), a three-hour meeting transcribes and summarizes in 2-3 minutes. On a CPU-only desktop, plan for 5-10 minutes. Older or slower hardware can take longer.

Can I share meeting transcripts with my team?

Yes. Transcripts export to markdown, PDF, or plain text. You can share the file or copy-paste the text into Slack, a doc, or email. If you deploy AnythingLLM to a server, multiple team members can access the same meeting library and create summaries together.

Does it work offline?

For transcription and summarization, yes. Once AnythingLLM and your LLM model are downloaded and running, you don't need an internet connection. Recording does not require internet. Uploading transcripts to Slack or exporting to cloud storage obviously does.

Is the output actually private?

The code is open source and auditable. Audio files are stored in AnythingLLM's local data directory (typically ~/.anythingllm on Mac/Linux, AppData\anythingllm on Windows). If you use a local LLM for summaries, nothing leaves your machine. If you mistakenly configure it to use an API LLM (OpenAI, Anthropic), that model will see the transcript. That's your choice, not AnythingLLM's default.

Can I edit the transcript after it's been created?

Yes. Open any meeting transcript and you can correct text, rename speakers, or delete segments. Changes are saved to your local database. If you've already created a summary, regenerate it after editing to reflect the changes.

Is there an API for integrating meetings with other tools?

AnythingLLM has an API for general chat and document operations. The Meeting Assistant API is limited, but the codebase is open so you can fork it or contribute back if you need custom integrations. For now, export-and-integrate (copy transcripts to Slack, Jira, etc.) is the intended workflow.

How far back can I search in old meetings?

Searchable history depends on your storage. AnythingLLM indexes every meeting you process, so if you import 100 old recordings, you can search all of them. The practical limit is disk space and how long your search queries take to run. Most users search the last 100 meetings without noticeable lag.

The Bottom Line

AnythingLLM's Meeting Assistant is a free, open-source alternative to Otter and Fireflies that runs on your own hardware and keeps your data private. It transcribes and summarizes accurately, integrates with your knowledge base so meetings feed into AI search, and eliminates recurring SaaS costs. Setup is fast, transcription quality is professional, and customizable templates make summaries actually useful instead of generic.

You should use it if you value privacy, want to cut meeting assistant costs to zero, or need to index meetings into a RAG system. You should skip it if your team requires seamless Salesforce integration or if running local infrastructure feels like overkill for your team size.

Ready to try it? Start with AnythingLLM's free desktop app and process one meeting to see if it fits your workflow. Or learn more about managed AnythingLLM hosting with Opsily, which handles infrastructure while keeping data in your control.