The AI Meeting Assistant Market Settles Into Clear Price Tiers

The AI Meeting Assistant Market Settles Into Clear Price Tiers

Fresh 2026 comparisons of Otter, Fireflies, Fathom, tl;dv, and Granola show the category has matured into predictable pricing bands. Entry paid plans cluster around $8–10 per user per month (Otter around $8.33–17 annualized, Fireflies $10–18), while Fathom has carved out a strong unlimited free tier for individuals before charging $15–19 for team features [1][2]. Granola sits at $14–18 with its bot-free, local-recording pitch, and tl;dv runs higher at $18-plus [3].

The free-tier fine print is where the real differentiation shows up: Fathom offers genuinely unlimited recording and transcription for individuals, Otter caps free usage at 300 minutes a month, and Fireflies leans on generous storage rather than unlimited minutes [2]. For teams evaluating tools, the decision increasingly comes down to CRM integration depth and whether a "bot joins your call" model or a quieter sidekick approach fits your culture.

It's a sign the category has moved past feature wars into a more mature, margin-conscious phase — good news for buyers, but a squeeze for vendors without a real differentiator beyond transcription.

Knowledge Graphs Take Center Stage as the Answer to AI's Memory Problem

A cluster of research and analysis this week converges on one theme: vector search alone isn't enough for agents or assistants that need to reason over long-term knowledge. InfoWorld's rundown of agentic memory systems, a new arXiv survey on graph-based agent memory, and a Forrester brief all point toward knowledge graphs — not flat embeddings — as the structure that lets AI systems track relationships, decisions, and dependencies over time [1][2][3].

The taxonomy emerging here splits memory into short-term/long-term and knowledge/experience categories, with techniques for extraction, storage, retrieval, and evolution of that knowledge as it accumulates. Forrester's framing is blunt: without semantics and ontologies, enterprise agents can't reliably reason — they just retrieve. Systems like Graphiti's temporal knowledge graphs are cited as examples of memory that evolves rather than just accumulates [2][3].

On X, the conversation has coalesced around a simple but sharp point: vector databases are good at "this sounds similar" but bad at "this depends on that." For any tool trying to make sense of months of meetings, that distinction is the whole ballgame.

Company Brain Goes Open Source, Showing the Blueprint for AI Teammates

Supermemory open-sourced Company Brain, a Slack-native AI teammate that ingests team conversations, remembers who owns what, answers questions from internal knowledge, and can take actions like opening issues in Linear [1]. It's notable both for the shift from paid product to open architecture, and for what it reveals about the plumbing required: multi-user Slack integration, persistent memory with provenance and permissions, and connectors into tools like Notion and Linear [1][2].

Related open-source efforts — temporal knowledge graph engines built explicitly as "organizational memory" — reinforce that the industry is converging on a pattern: ingest from meetings, Slack, and docs, then turn that into executable context an agent can act on, not just search [2][3]. The founder's public write-up of the full architecture has drawn attention as a rare, complete blueprint rather than a marketing teaser.

This matters beyond Slack: it's essentially the same problem meeting-intelligence tools solve, applied to chat instead of audio. The permissions and provenance layer — knowing who said what, and who's allowed to see it — is the hard part everyone is now building in public.

What This Means For Your Meetings

Put these four stories together and a pattern emerges: the infrastructure for turning conversation into durable, queryable knowledge is rapidly commoditizing. Open-weight diarization means any vendor can now get accurate speaker identification without building it from scratch. Open-sourced "company brain" architectures show the same is happening for memory and retrieval. The barrier to entry for basic meeting transcription is falling fast — which means the value is shifting entirely to what happens after the transcript.

That's exactly where knowledge graphs come in. A flat transcript, even a perfectly diarized one, is just a record of what was said. The research this week on graph-based agent memory makes the case clearly: relationships between decisions, owners, and follow-ups need structure to be retrievable months later — "what did we decide about the Helsinki rollout in March, and who owns it now" is a graph query, not a keyword search. This is precisely the gap between a transcription tool and a genuine personal knowledge base.

The pricing comparisons are a useful reality check too: most of the market is still selling minutes transcribed, not knowledge retained. As diarization and basic capture become table stakes — free, open, commoditized — the tools that win will be the ones that build durable, speaker-aware knowledge graphs across your entire meeting history, not just per-call summaries.

Key takeaway: Capturing a meeting is becoming trivial; understanding it — who said what, how it connects to everything before it, and what to do next — is where the real product now lives.

Sources

1. https://www.marktechpost.com/2026/09/23/nvidia-releases-nemotron-3-diarization/
2. https://huggingface.co/blog/nvidia/nemotron-diarization
3. https://huggingface.blog/blog/nvidia-nemotron-diarization/
4. https://wirecraft.ai/best-ai-meeting-note-takers-2026/
5. https://dupple.com/learn/best-free-ai-meeting-assistants
6. https://bestautomationtools.ai/best-ai-meeting-assistants/
7. https://www.infoworld.com/article/4192397/four-agentic-ai-memory-systems-for-smarter-llms.html
8. https://arxiv.org/abs/2602.05665
9. https://www.forrester.com/blogs/build-meaning-before-machines-why-semantics-ontologies-and-knowledge-graphs-matter-for-agentic-ai/
10. https://github.com/supermemoryai/company-brain
11. https://github.com/caelstewart/company-brain
12. https://github.com/garrytan/gbrain/blob/master/docs/tutorials/company-brain.md