I usually just ask the agent to create a summary document in markdown format... I'll sometimes play out what I want to do on my phone in conversational modes, then take that output and start with it against a coding tool to continue/start actual work on things.
I tend to spend a lot more time planning than what it takes to execute/build what I'm after.. then more time reviewing again than the produce step(s) and iterate.
For that matter, I'll track what I'm wanting and what has been done via TODO.md file in project(s).
I'm not sure I would ever use/trust a service like this myself. I barely trust the AI providers as it is.
This is a paid, more complex version alternative to just telling your agent to "update the agents.md, plan.md and decisions.md for future agents after you finish and test your changes"?
I use two AI CLIs every day (one for main tasks and the other for review), and context migration is a real pain point.
My current workaround is to write the context as a Markdown file and save it to disk, then let both of two CLI to read. I'd like to ask you, what are the advantages of this tool compared to using a file as a bridge?
Very interesting. This is the #1 thing LLMs don’t have and should: discrete data handles. Like literally you can’t save & manage a todo list in the default Claude or ChatGPT web or iOS apps. So this is like a pastebin that works cross chat I assume
Others have already pointed out the absurdity, but just to put it across. Why not just ask first agent to update AGENTS.md adding context and next steps, and ask the next agent to continue from there?
It’s not a bad idea in itself, but as others have already said, it can be handled more securely using a local file. Especially now, with all the discussion about sovereignty and local hosting, a service like this is rather counterproductive.
But perhaps this could be used to build a local solution with a little more added value. It might be interesting to use the context to determine which AI solves the problem best or most cost-effectively.
Either I'm an idiot or am missing something entirely (maybe a demo video would help?)
I've got to send it to your server so you can spit it back when I say a fancy word? What's worse, I pay you to just store/retrieve text? Why wouldn't I just write to a file?
I'm glad you posted this before I took the time to review the link; saved me time. I'm certainly interested in a feature like this that's all local in my file system as I'm sure others are as well.
I like the name! I'd consider a self-hosted and open-source version as your "free" plan then a cloud-hosted paid version, and for $10+ if your goal is to make a profit vs. run a hobby project.
Why? "Summarize in detail what we're doing here and what we have done/learnt so far so another agent can continue" usually works and does not burn the full context length of tokens (uncached) in the new session.
What was your initial problem? I usually work with contexts where they were created, like code session in Cursor / chat in Claude app. When do you have the need to often transfer the history?
I usually just ask the agent to create a summary document in markdown format... I'll sometimes play out what I want to do on my phone in conversational modes, then take that output and start with it against a coding tool to continue/start actual work on things.
I tend to spend a lot more time planning than what it takes to execute/build what I'm after.. then more time reviewing again than the produce step(s) and iterate.
For that matter, I'll track what I'm wanting and what has been done via TODO.md file in project(s).
I'm not sure I would ever use/trust a service like this myself. I barely trust the AI providers as it is.
This is a paid, more complex version alternative to just telling your agent to "update the agents.md, plan.md and decisions.md for future agents after you finish and test your changes"?
I use two AI CLIs every day (one for main tasks and the other for review), and context migration is a real pain point.
My current workaround is to write the context as a Markdown file and save it to disk, then let both of two CLI to read. I'd like to ask you, what are the advantages of this tool compared to using a file as a bridge?
This is such a great name and domain for this purpose.
I don’t think I’d like using a third party server for this, but a skill that just handles it locally or something would be useful.
Very interesting. This is the #1 thing LLMs don’t have and should: discrete data handles. Like literally you can’t save & manage a todo list in the default Claude or ChatGPT web or iOS apps. So this is like a pastebin that works cross chat I assume
Ask your LLM in session to create a summary.md file... you can carry that into other sessions with different/new contexts.
Others have already pointed out the absurdity, but just to put it across. Why not just ask first agent to update AGENTS.md adding context and next steps, and ask the next agent to continue from there?
It’s not a bad idea in itself, but as others have already said, it can be handled more securely using a local file. Especially now, with all the discussion about sovereignty and local hosting, a service like this is rather counterproductive.
But perhaps this could be used to build a local solution with a little more added value. It might be interesting to use the context to determine which AI solves the problem best or most cost-effectively.
Either I'm an idiot or am missing something entirely (maybe a demo video would help?)
I've got to send it to your server so you can spit it back when I say a fancy word? What's worse, I pay you to just store/retrieve text? Why wouldn't I just write to a file?
I'm glad you posted this before I took the time to review the link; saved me time. I'm certainly interested in a feature like this that's all local in my file system as I'm sure others are as well.
I like the name! I'd consider a self-hosted and open-source version as your "free" plan then a cloud-hosted paid version, and for $10+ if your goal is to make a profit vs. run a hobby project.
This is dangerous.
You're sending your context to someone's relay in plain text and loading it in plain text.
Don't worry, they pinky promise not to save anything.
Isn't saving it kind of the point?
Why? "Summarize in detail what we're doing here and what we have done/learnt so far so another agent can continue" usually works and does not burn the full context length of tokens (uncached) in the new session.
What was your initial problem? I usually work with contexts where they were created, like code session in Cursor / chat in Claude app. When do you have the need to often transfer the history?