Please let me know if the image doesn't provide sufficient context and what other info you'd need to help. Thank you! (I'll be gone much of today but will check back as possible.) |
Replies: 4 comments 4 replies
|
Hey @ckbarlow , LFM 2.5B might be too small to be great with tools. Generally, smaller models are not a great daily assistants. They are great at specific tasks with very limited tools available to them. You could have multiple agents with individual configurations that makes smaller models more useful. If you would like it to do specific work, I would suggest doing the following:
|
|
So this is what you’d recommend for something like creating a reminder per email from a specific sender? |
|
Can you walk me through your ideal flow?
|
|
Edited to make less obnoxiously long Yes:
From there, when I want to do one of the listed tasks, I click the mail link. Then I trigger one of many, many Keyboard Maestro macros I use to access the task's files I need and set them up the way I need. This was just a test case to start learning Osaurus. I don’t know if you would consider Taskade anywhere close to what you’re doing, but I tried that for several months and got annoyed with it. Context: I’m a composer. I'm not permitted to, and don't want to, use AI for any creative work. With Taskade, I did things like create an agent per piece of gear in my studio and train it on the user manual. Still, hallucinations when I would ask how-to questions. More of a waste of time than anything, unfortunately. I've also used Claude locally with Notes extensions to organize my massive collection of music production notes. Hoping Osaurus will prove more useful. Thanks again for your time. |
|
Update: I have
Me: Create a Reminder based on the currently displayed email Agent: Error: Ran out of system resources while running this model. Free memory, unload other models, or choose a smaller/more-quantized model, then try again. Maybe 24GB RAM isn't enough. |
|
Hey @ckbarlow, sorry to say 24GB RAM is a bit small to handle agentic tasks at this time. We're researching ways to improve that, but for now, I would say it's a tight fit. For your specific use case though, we do have Resend plugin (https://github.com/osaurus-ai/osaurus-resend), but it requires setting up a custom email for your agent (doesn't read through your mailbox). This will work better with smaller models. Instead of LFM2.5, try gemma-4-e4b-it-4bit which is slightly less capable but with enough "harness tuning" you can get this to work no problem. If you want to try using a remote frontier model (recently added Grok provider support), here's what I would do to achieve your setup:
I hope this helps! |
|
Thank you! |
|
OK, cool. Perhaps add RAM recommendations to the System Requirements <https://docs.osaurus.ai/installation#system-requirements>?
I’m on an M4 MacBook Air, 2TB drive, 24GB RAM. For extensive Logic Pro X use and video editing, it’s been excellent. (I’ve always had Pros until this M4 Air - it’s amazing!)
I guess local agentic work is a new level of heavy.
Thank you for your help. I’ll probably uninstall, finish my current composing projects and then update to 26 / Foundation, see how that goes.
Thanks again,
CK
… On Jun 7, 2026, at 9:31 AM, tpae ***@***.***> wrote:
Hey @ckbarlow <https://github.com/ckbarlow>, sorry to say 24GB RAM is a bit small to handle agentic tasks at this time. We're researching ways to improve that, but for now, I would say it's a tight fit.
For your specific use case though, we do have Resend plugin (https://github.com/osaurus-ai/osaurus-resend), but it requires setting up a custom email for your agent (doesn't read through your mailbox). This will work better with smaller models. Instead of LFM2.5, try gemma-4-e4b-it-4bit which is slightly less capable but with enough "harness tuning" you can get this to work no problem.
If you want to try using a remote frontier model (recently added Grok provider support), here's what I would do to achieve your setup:
Enable Agent Database (Agents > Agent > Configure > Database)
Enable Self Scheduling (Agents > Agent > Configure > Self-scheduling)
Enable mail and reminder tools.
Initial provision prompt: ask agent your intention, and you want to keep track of each email received, keep a separate status. ask to use the agent database to make sure we keep track.
Ask agent to create a schedule to fetch and add to reminders (give it some time interval)
I hope this helps!
—
Reply to this email directly, view it on GitHub <#1383?email_source=notifications&email_token=BOOKR4Q7NCLQXKFAKMDVJSD46VVEJA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNZSGA4TQNBRUZZGKYLTN5XKO3LFNZ2GS33OUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#discussioncomment-17209841>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BOOKR4WCBHTKTH2IL7V4PRD46VVEJAVCNFSM6AAAAACZ5ABDM2VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTOMRQHE4DIMI>.
Triage notifications, keep track of coding agent tasks and review pull requests on the go with GitHub Mobile for iOS <https://github.com/notifications/mobile/ios/BOOKR4T26UHGM2EIARULZOL46VVEJA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNZSGA4TQNBRUZZGKYLTN5XKO3LFNZ2GS33OUVSXMZLOOSVGM33PORSXEX3JN5ZQ> and Android <https://github.com/notifications/mobile/android/BOOKR4X6ORMKVOTVPXTU4TT46VVEJA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNZSGA4TQNBRUZZGKYLTN5XKO3LFNZ2GS33OUVSXMZLOOSXGM33PORSXEX3BNZSHE33JMQ>. Download it today!
You are receiving this because you were mentioned.
|
Hey @ckbarlow, sorry to say 24GB RAM is a bit small to handle agentic tasks at this time. We're researching ways to improve that, but for now, I would say it's a tight fit.
For your specific use case though, we do have Resend plugin (https://github.com/osaurus-ai/osaurus-resend), but it requires setting up a custom email for your agent (doesn't read through your mailbox). This will work better with smaller models. Instead of LFM2.5, try gemma-4-e4b-it-4bit which is slightly less capable but with enough "harness tuning" you can get this to work no problem.
If you want to try using a remote frontier model (recently added Grok provider support), here's what I would do to achieve your setup: