Conversation → Note — every AI chat, saved as real knowledge.

ChatGPT's own export gives you a JSON dump and a seven-day wait. Knovya gives you a structured note — with decisions, action items, and citations auto-extracted, in one click. Three capture methods (native chat, paste, MCP roundtrip), four AI clients, and a closed loop with AI Memory. Free includes 50 capture credits per month; Pro raises it to 500 and unlocks the MCP roundtrip.

Capture methods
3
AI clients supported
4
Auto-extracted parts
5
Conversation → Note
Experiment 01 · The Lab

Pick a source. Watch it become a note.

Three capture methods, one structured note. Switch the source on the left — the right panel rebuilds, with decisions, actions, and citations extracted as it streams.

Source
In Multi-turn AI chat
Native
Help me plan a Q3 launch strategy for our SaaS product.
Got it. Who is the target audience for this launch?
SaaS founders running $100K–$1M ARR. Series A folks mostly.
And the timeline — full quarter, or front-loaded?
Front-loaded. Big push in week 1–4. Then monitor.
One more — paid channels, organic, or both?
Organic-first. Paid only as a top-up if NRR slips.
Got enough. Writing the plan now.
Pasted from chatgpt.com · Mar 18 · GPT-5.4 ThinkingYou: Help me plan a Q3 launch strategy for our SaaS product. ChatGPT: Sure — could you tell me more about the product, audience, and timeline? You: SaaS for SaaS founders, $100K–$1M ARR. Q3, front-loaded, organic-first. We need to push retention over growth this quarter — NRR target is 110%. ChatGPT: Here is a draft plan: Week 1–2 launch retention dashboard; Week 3–4 founder-led content series; Week 5–8 community AMA + cohort program. Decision: prioritize NRR > new logo growth. Action items: draft retention OKR, set NRR target by W2, ship dashboard W1. Source: prior Q2 retention report.
// Claude Desktop · MCP roundtrip { "tool": "knovya_write", "args": { "title": "Q3 launch strategy", "folder": "Plans / Q3", "source": "claude-conversation", "transcript": "…7 turns…", "tags": ["launch", "q3", "retention"], "extract": [ "decisions", "actions", "sources" ] } } // Knovya replies with note_id + structured plan
Out Knovya note
Auto-structured
Auto-extracted from this conversation
0 decisions 0 action items 0 sources
Free includes 50 capture credits per month — paste and extension methods. Pro raises this to 500 credits and unlocks the MCP roundtrip → See Pro
The anatomy · how Knovya captures

Three methods, one structured note.

Each path captures a different shape of conversation. All three converge on the same Knovya note — searchable, linkable, and ready to act on.

Method I Native chat
in Knovya
01
Multi-turn information gathering Knovya AI asks for the audience, the constraints, the success criteria — until the picture is clear. The conversation is the brief; you do not have to write one.
02
Readiness check When enough context has been gathered, Knovya offers to write — you can also tell it "write now" earlier, and it will warn if anything material is still missing.
03
Streamed structure The note arrives section by section as the model writes it — headings, lists, quotes, callouts — already block-shaped, never a wall of prose to clean up.
Method II Paste a transcript
from anywhere
04
Smart paste handler Drop a ChatGPT, Claude, Gemini, or Perplexity transcript into a Knovya note. The handler folds it into a collapsed source card — never floods the editor — and asks if it should structure it.
05
Source auto-detection Knovya recognises the export shape — ChatGPT's "You:" / "ChatGPT:" markers, Claude's role tags, Gemini's share format — and tags the note with the originating client.
06
Cleanup & restructure Filler is removed, duplicates merged, headings inferred. The chat becomes a document — turn order preserved, but no longer formatted like one.
Method III MCP roundtrip
the magic
07
The AI saves itself Tell Claude or ChatGPT "save this conversation to my Knovya plans folder." Through MCP, the client calls knovya_write with the transcript — no copy-paste, no extension.
08
Round-trip preserved Knovya replies with the new note's id and a structured summary. Your AI sees the saved note immediately — and can keep working from it in the same chat.
09
Folder & tag intent The same MCP call accepts a folder, tags, and metadata — so the AI can place the note where it belongs, not just dump it into an inbox.
Across all three Auto-extraction
always on
10
Decisions, lifted out Whenever the conversation says "let us go with X," Knovya pulls that line out as a Decision block — one of the three things you usually wanted from the chat in the first place.
11
Action items, made checkable Things like "I will draft the OKR by Friday" become checklist items at the top of the note — owner-tagged when the conversation makes it obvious.
12
Sources & citations URLs that came up are footnoted; documents you paraphrased are linked back; tools the AI called are listed. The note is auditable, not just readable.

Your most useful AI conversation just disappeared.
And the wait is seven days.

The chat where you finally figured out the architecture. The thread where the strategy clicked. The voice memo of the meeting that actually mattered. All trapped in tabs that close, accounts that get cancelled, models that get retired.

ChatGPT's official export is a ZIP of raw JSON, delivered up to seven days later. Browser extensions ship the chat as a PDF or a Notion page — clean format, but the decisions stay buried in chat bubbles. The action items never become actionable.

The cost
One ChatGPT export tool alone has 18,000+ active users archiving conversations they will never re-read.
The fix
Stop archiving the chat. Extract the knowledge.
The lineage

From the Memex to the loop.

Capturing what we want to remember is an old idea. Each generation found one new way to keep it. Conversation → Note finishes the chain.

  1. 1945
    Vannevar Bush — the Memex "As We May Think." A desk that remembers every reference you ever followed — the first vision of associative knowledge capture. The Atlantic · essay
  2. 1987
    HyperCard — linked stacks Bill Atkinson proved capture was a UI problem, not a storage one. A note pointed at another note, and a generation learned to think in links. Apple · Macintosh
  3. 2009
    The web clipper Evernote's clip-and-save button made the browser part of the notebook. Articles came in. Conversations did not. Evernote · browser extension
  4. 2022
    ChatGPT — the unsavable conversation Suddenly the most useful thinking was happening inside chat windows. Native export shipped 18 months later, as a ZIP. Extensions filled the gap by saving the form, never the substance. OpenAI · production
  5. 2026
    Knovya — Conversation → Note Three capture methods, structured by AI, surfaced through MCP. The decisions, the actions, the citations — kept. The loop closes. ★ Knovya · production
First mover

Every other tool ships the conversation. Knovya ships the knowledge.

ChatGPT Toolbox saves a PDF. Claude → Notion lifts the messages into a Notion page. Mem.ai chats with notes you already saved. Basic Memory imports the JSON dump and sits it in a graph. Nobody auto-extracts the decisions, the actions, the sources — and feeds the result back to your AI through MCP.

  • ChatGPT Toolbox PDF · MD · JSON
  • Claude → Notion page transit
  • ChatGPT Projects project-locked
  • Mem.ai chat with own notes
  • Basic Memory JSON import
  • ★ Knovya structured note · auto-extracted · MCP loop
Surfaces

Four ways it lives in your day.

Inside Knovya, in your browser, inside Claude, and back again. The capture method changes; the structured note does not.

In-app capture native

Open Knovya, hit ⌘N, choose your capture method. The same picker you saw in The Lab, sitting inside the editor — one keyboard shortcut from any note.

Browser extension on-page

A button that lives on chatgpt.com and claude.ai. Click it on any conversation tab and Knovya saves the structured note — folder and tags pre-filled from the page context.

MCP roundtrip agent-callable

When Claude or Cursor calls knovya_write, the conversation saves itself. No extension, no copy-paste — the AI puts its own thinking into your knowledge base.

The closed loop memory layer

The saved note enters AI Memory. Your next ChatGPT or Claude session reads it through MCP — and picks up where the last one left off, without a single re-explanation.

Bonded with

Cv composes with the rest of Group I.

01
MCP
The protocol behind the roundtrip — Claude, Cursor, ChatGPT calling Knovya directly.
02
AI Memory
Where the saved conversation lives. The next session reads what the last one decided.
06
Web Research
Same pattern, different source — synthesise URLs into a structured note with citations.
10
Meeting Notes
Transcript in. Structured minutes out. The same idea, played on a different input.
Frequently asked

A few honest answers.

How do I save a ChatGPT conversation?
There are three ways. One — paste the conversation into Knovya: the smart paste handler detects ChatGPT and Claude transcripts and structures them into a note. Two — install the Knovya browser extension and click Save on any chatgpt.com or claude.ai page. Three — connect Knovya to your AI client through MCP, and the conversation saves itself when you ask it to.
Can I save a Claude conversation to Knovya?
Yes. The same three methods work for Claude. The smart paste handler recognises Claude's markdown export, the browser extension lives on claude.ai, and the MCP roundtrip is the most direct of all — Claude itself can call knovya_write to save the conversation as a note.
Does Knovya export ChatGPT to PDF or Markdown?
Once a conversation is saved as a Knovya note, you can export it to Markdown, PDF, DOCX, HTML, or JSON from the editor — the same way as any other note. The difference: Knovya stores the structured note, not the raw chat, so the exported file already has headings, decisions, and citations laid out.
Are saved AI conversations searchable?
Yes. Every conversation that becomes a Knovya note is indexed by the same hybrid search that runs on the rest of your knowledge base — full-text plus semantic, reranked by NoteRank. Ask a question and the relevant chat surfaces alongside your other notes.
Will Knovya summarise the chat?
It does more than summarise. As the conversation streams, Knovya extracts decisions, action items, sources, and attendees into their own blocks — so the note is ready to act on, not just ready to skim.
Does this work with Cursor, Gemini, and Perplexity?
Yes. Any AI client that speaks MCP can save a conversation directly. For clients without MCP yet, the paste method works on every transcript format we have tested — Cursor's chat export, Gemini's share-link, Perplexity's thread. The browser extension currently supports ChatGPT and Claude, with Cursor and Gemini in progress.
How is this different from ChatGPT Toolbox or Claude-to-Notion extensions?
Those extensions ship the conversation. Knovya ships the knowledge inside it. ChatGPT Toolbox saves a PDF of the chat. Claude-to-Notion lifts the messages into a Notion page. Knovya parses the conversation, extracts the decisions and actions, links the sources, and feeds the result back to your AI through MCP — so the next session knows what was decided last time.
How many conversations can I save on the Free plan?
Free includes 50 conversion credits per month, which is enough to capture roughly 25 to 30 medium-length AI conversations. Pro raises this to 500 credits per month, unlocks the MCP roundtrip method, and removes the per-month cap on browser-extension captures.

Stop losing AI conversations. Start the loop.

Free includes 50 capture credits per month — paste and extension. Pro raises it to 500 and unlocks the MCP roundtrip on Claude, Cursor, and ChatGPT.

Start free See Pro
element 09 · Group I — AI