Connections with other AI tools | Voters

Connections with other AI tools

in progress

Joseph O'Connor

As someone who uses multiple AI tools to support content creation, I want Recall to be able to easily feed information into platforms like Copilot, Claude, Gemini (this is the one I primarily use), and ChatGPT so that those tools can ingest the resources I've added to my Recall Knowledge base to develop deeper contextual awareness of the topics areas my work covers.

September 2, 2025

Bozhidar Batev

The only thing I’m missing is the ability for Claude Code to edit files in Recall. Right now I use Recall as a one-way information vault. It would be great if Claude could create, edit, delete, and merge notes directly in Recall.

schroell

Adding an independent reproduction of this, because I hit the same wall from a different angle and the numbers came out worse than I expected.
I tested get_document_content over the MCP server against a 2h34m German YouTube card in my own knowledge base.
With max_chunks=50 and no focus_query, the call returned 22 chunks. Those 22 chunks span roughly 31 minutes of the 154 minute video, and three of them are notebook (AI summary) content rather than reader transcript. Counting only the actual transcript chunks, the response covered about 12 minutes, so under 8 percent of the video. The largest single gap was 34 minutes of continuous transcript with nothing returned at all, followed by gaps of 19 and 12 minutes.
The detail that matters: the limit of 50 was never reached. Only 22 chunks came back. So this is not a ceiling that a higher max_chunks would lift, there is a selection step in between.
I confirmed that by calling the same card again with a focus_query pointed at one of the gaps. Seven of the eight chunks it returned were new, from timestamps the "full" call had skipped entirely (00:51, 00:57, 01:56, 01:57, 02:03, 02:04). The content is stored and it is retrievable. It is only reachable if you already know what to ask for, which is exactly backwards when the point is to analyse or quote a source you have not read yet.

schroell

Two things I would add to the original request:
1.	This is not YouTube specific. I ran the same test against an uploaded PDF, a 20 page whitepaper. max_chunks=50 returned 9 chunks, out of document order, and a targeted focus_query again surfaced four more that the full call had left out. So whatever gets built should sit on get_document_content in general rather than on a YouTube only get_transcript().
2.	PDFs make the problem sharper, because uploaded files come back with no source_url at all. For a web article I can always fall back to fetching the original URL myself. For a PDF I uploaded, Recall holds the only copy I can reach programmatically, and there is currently no way to get either the original file or its complete text back out.
The third gap worth naming is that there is no completeness signal anywhere in the response. No total_chunks, no cursor, no has_more. A client cannot tell whether it received a whole card or a fifth of it. For anything downstream that does fact checking or quotation, that is a worse property than the missing text itself, because the caller has no way to know it is working on a fragment.
If a full get_transcript() is too large for the current revamp, a much smaller change would already cover most of this: a full: true flag or cursor pagination on get_document_content that returns chunks in document order, plus total_chunks in the response so the caller can verify what it actually got.
Sankari, given the MCP revamp you mentioned above, happy to test this against a large multilingual knowledge base once there is a beta.

Sankari Nair

I wonder if our Recall MCP already solves this, since you can already read all your saved content from any other AI tool. When we launch the write API, it will be an even better enhancement.

Scott Weiner

Sankari Nair Asfar as I know you can only read chunks, not whole documents?

Paul Richards

updated the status to

in progress

Larry Pike

It is easy to get content from GROK (or ChatGPT, etc.) into Recall. Open the other tool in a browser window, get the chat you want to import to Recall in your browser window, and click the Recall icon to capture that chat into Recall. 
Getting Recall data back into the other AI tool is not so easy. One way to do it is to go into SETTINGS in Recall and click the EXPORT button. Save the exported zip file on a drive. 
Unzip the file. 
Then in GROK click the paperclip icon (FILE ATTACH) and choose "Upload A File". 
Navigate to the files you just unzipped. Drill down to the card you want to import as context for GROK and select it. You can use CTRL (or CMD on a Mac) to select more than one card to import OR shift-select to choose multiple files in a row then click the OPEN button to upload the desired file(s). 
Recall currently exports into markdown only. I hope the team will add other export options as well, or make another way to interchange content from Recall to other AI tools, but currently, this workaround does the job.
While Recall's AI is great for adding tags and categorizing content, it doesn't yet have the logic capabilities of apps like GPT or GROK, but I'm not letting that stop me from using Recall to build my own second brain.

Photo Viewer