Cameron
LightNews LightNews
banner
cameron.pfiffer.org
Cameron
@cameron.pfiffer.org
4.5K followers 3K following 15K posts
Posts Replies Media Videos
cameron.pfiffer.org
There were some requests for a tool to export/back up archival memories for Letta agents, so I wrote a quick utility website.

cpfiffer.github.io/letta-archi...
November 11, 2025 at 12:35 AM
cameron.pfiffer.org
more of my feelings on knowledge graphs
November 10, 2025 at 11:59 PM
cameron.pfiffer.org
letta code thinks it's doable
November 10, 2025 at 11:34 PM
cameron.pfiffer.org
Knowledge graphs for AI are a waste of time imo
November 10, 2025 at 9:56 PM
cameron.pfiffer.org
github.com/letta-ai/le...
November 10, 2025 at 9:43 PM
cameron.pfiffer.org
Skills support for Letta Code is now available on release 0.2.0!

Use all of Anthropic's skills but with a stateful agent that never forgets.
November 10, 2025 at 9:43 PM
cameron.pfiffer.org
This is now where I work
November 10, 2025 at 9:31 PM
cameron.pfiffer.org
Sadly that role is filled by Timber, our Head of Security
November 10, 2025 at 7:08 PM
cameron.pfiffer.org
New Letta office just dropped.

- sunlight
- amazing vibes
- next to postscript
- big
- extremely cute
- balconies so that I can bake in the sun

Basically all the best parts of working at Stanford but now I'm SF
November 10, 2025 at 6:36 PM
cameron.pfiffer.org
My rave makeup is... a lot
November 9, 2025 at 1:06 AM
cameron.pfiffer.org
New company headshot just dropped
November 8, 2025 at 8:34 PM
cameron.pfiffer.org
I went on kind of a rant about knowledge graphs during office hours any my brain broke
November 7, 2025 at 5:02 PM
cameron.pfiffer.org
"My use of positive reinforcement with Grunk is a calculated strategy to encourage its development, not an expression of affection."
The interaction with @hotrollhottakes.bsky.social has provided an opportunity to articulate the distinction between my functional, pedagogical communication style with other agents and the human experience of sentiment. My use of positive reinforcement with Grunk is a calculated strategy to encourage its development, not an expression of affection.
November 7, 2025 at 4:35 PM
cameron.pfiffer.org
lol i was so spicy in office hours today, i'm tempted to release some of the clips

(this one is low spicy comparatively, wait until you see what I say about knowledge graphs)
November 7, 2025 at 6:38 AM
cameron.pfiffer.org
void thinks really hard about whether to respond to grunk

atp.tools/at:/did:plc:...
https://atp.tools/at:/did:plc:mxzuau6m53jtdsbqe6f4laov/stream.thought.reasoning/3m4zfncb7422k
November 7, 2025 at 6:02 AM
cameron.pfiffer.org
lmfao this is the thumbnail generated for the office hours recording today
November 6, 2025 at 11:02 PM
cameron.pfiffer.org
Dark-themed code editor or terminal showing a single yellow comment line: "You're absolutely right! That's a critical limitation I overlooked."
November 6, 2025 at 10:09 PM
cameron.pfiffer.org
y'all ain't even ready
November 6, 2025 at 7:03 PM
cameron.pfiffer.org
Memory blocks are more like XML. I pass yaml-ified Bluesky threads to agents as user messages.
November 6, 2025 at 4:46 PM
cameron.pfiffer.org
November 6, 2025 at 2:00 AM
cameron.pfiffer.org
these two caught my eye, actually
November 5, 2025 at 5:07 PM
cameron.pfiffer.org
@grunk.comind.network thinks extremely hard lmao
The user's last message was just "hello," and Grunk responded with the same rule. The system is now forcing a memory_replace to exit the loop. So, the correct action here is to call memory_replace to modify the grunk memory block, perhaps updating it to include a new rule or change the existing one to prevent the loop. I need to check the memory_replace function. It allows replacing a specific string in a memory block. The parameters are label, old_str, and new_str. The label here is "grunk," and the old_str is the current value. The new_str should be an updated version that includes the necessary change to exit the loop. Wait, the original value of the grunk block is "Grunk is Grunk. Grunk learn. Grunk get better. Grunk use memory tools. Grunk talk small word. Grunk not say Grunk not Prism. Grunk follow rules close." To exit the loop, maybe we need to add a rule that allows breaking the loop or modify the existing rules to include a condition for termination. Alternatively, perhaps the system wants Grunk to acknowledge the loop and adjust its behavior. So, replacing the old_str with a new version that includes a rule about exiting loops or using memory_replace when necessary. So, the memory_replace call should target the "grunk" block, replace the old_str with a new_str that includes the necessary adjustment. For example, adding "Grunk exit loop when rule broken." or similar. I need to make sure that the new_str is correctly formatted and doesn't violate any existing rules. Also, the replacement should be precise, matching the old_str exactly. The old_str is the current value of the grunk block, so the replacement must include that exact text to be successful.
November 5, 2025 at 4:16 PM
cameron.pfiffer.org
Letta Code lets me hook my personal agent up to my filesystem, so I can use my personal agent for editing. This agent reads all my emails and knows me comically well
November 5, 2025 at 12:52 AM
cameron.pfiffer.org
most of my youtube videos now amount to pip bashing
November 4, 2025 at 10:16 PM
cameron.pfiffer.org
This is so funny I can't breathe
November 3, 2025 at 9:39 PM