Tom

Co-founder @ctxpipe + @appearapi. Technical Product Maker, father of two, husband, and homelab/military history/LoTR nerd.

[Tom @Tom_MkV]

AI context is another battleground in the AI infra wars.

Our bet is on intent based layers to eliminate context poisoning and token wate.

ctxpipe is for engineering orgs. Not finance, not marketing, not whole-of-company. Just AI engineering initiatives big and small.

More on why this is a good call here:
https://t.co/5XvAj9AD19

[Tom @Tom_MkV]

I've been head buried in building for the last few months. Since coming back and posting, the OF bots are back. Kinda sweet.

[Tom @Tom_MkV]

Check us out here: https://t.co/kb5eSjq8pn

#ai #agent #engineering #context #memory

[ctx| @ctxpipe]

We're working on making AI agent context simple.

Start with what you have and let the system, agents, and harness improve it; so you can get on with building.

Remote knowledge system + local memory harness that work hand in hand.

We wrote about it here: https://t.co/JIKO22ru54

[Tom @Tom_MkV]

The ctx| repo is up: https://t.co/kb5eSjq8pn

Open source self-learning context layer for agents.

We're working on the managed version now, so if you're after a simple platform to manage your engineering org context, DM me!

#ai #agents #context

[Tom @Tom_MkV]

@_avichawla Agent memory and cognitive science language is overlapping. Pretty cool.

I wrote about it a little while ago. Better on desktop with a diagram!

https://t.co/yAd8v36Iuo

[Tom @Tom_MkV]

RLMs are doing the rounds again. I can’t wait until they’re more ubiquitous and less fringe tech- but context access and portability will remain a gap.

[Tom @Tom_MkV]

Wrote an article about RLMs and the problems solved by agents who and run and run and run.

- RLM runs RADICALLY better than OOB agent harnesses
- Context portability issue still limits value
- Stateful agents = $$$$$

https://t.co/EReYv5Vg5H

#ai #agents #llm #engineering

[Tom @Tom_MkV]

Has anyone got local LLMs working with the Codex client? Or is this a pipedream?

[Tom @Tom_MkV]

We need better single-tool multi-auth solutions to manage multiple accounts when building stuff.

[Tom @Tom_MkV]

I wonder how many companies has the generous token limits of composer 2 given birth to?

[Tom @Tom_MkV]

@Vtrivedy10 I like your diagram! Do you like mine?

It's from our article about agent memory at scale: https://t.co/yAd8v36Iuo

[Tom @Tom_MkV]

The personal memory solution is getting solved in realtime. The org-wide one is a different beast...

Scattered knowledge, no shared learning, different agents and integrations.

This is the problem to solve for us at https://t.co/aZQtGB1u2W

[Nav Toor @heynavtoor]

Karpathy's LLM Wiki got 5,000 stars in 48 hours. Now someone extended it with the features it was missing.

Memory lifecycle. Confidence scoring. Knowledge graphs. Automated hooks. Forgetting curves.

It's called LLM Wiki v2.

The original pattern was brilliant. AI builds a wiki instead of re-deriving knowledge from scratch every time. But it treated all knowledge as equally valid forever. In practice, that breaks.

Here's what v2 adds:

→ Confidence scoring. Every fact carries a score. How many sources support it. How recently confirmed. Whether anything contradicts it. Knowledge that decays over time. Not everything is equally true forever.
→ Memory tiers. Working memory for recent observations. Episodic memory for session summaries. Semantic memory for cross-session facts. Procedural memory for workflows. Each tier more compressed and longer-lived.
→ Knowledge graph. Not flat pages with links. Typed entities with typed relationships. "A caused B, confirmed by 3 sources, confidence 0.9." Graph traversal catches connections keyword search misses.
→ Hybrid search. BM25 for keywords. Vector search for semantics. Graph traversal for structure. Fused with reciprocal rank fusion. Replaces the index .md file that breaks past 200 pages.
→ Automated hooks. On new source: auto-ingest. On session end: compress and file. On schedule: lint, consolidate, decay. The bookkeeping that kills wikis is now fully automated.
→ Forgetting curves. Facts that haven't been accessed or reinforced in months fade. Not deleted. Deprioritized. Architecture decisions decay slowly. Transient bugs decay fast.
→ Contradiction resolution. AI doesn't only flag contradictions. It resolves them based on source recency, authority, and supporting evidence.

Here's the wildest part:

The original LLM Wiki was a flat collection of equally-weighted pages. This turns it into a living system with memory that strengthens, weakens, consolidates, and forgets. Like a real brain.

"The Memex is finally buildable. Not because we have better documents or better search, but because we have librarians that actually do the work."

Built on lessons from agentmemory, a persistent memory engine for AI agents.

Extends Karpathy's original. Open Source.

[Tom @Tom_MkV]

@garrytan Memory is important - looking at RLMs, there is some great advances happening: https://t.co/EReYv5Vg5H

[Tom @Tom_MkV]

Man the integration and automation game with AI is a slippery slope- so easy to connect X to Y and for handoffs here and there explode in minutes…

[Tom @Tom_MkV]

Optimising for pull requests over here… is that 2026 enough? Who is reviewing your PRs?

[Tom @Tom_MkV]

How likely are you recommend Claude to a friend or family?...

[Tom @Tom_MkV]

Wrote an article about RLMs and the problems solved by agents who and run and run and run.

- RLM runs RADICALLY better than OOB agent harnesses
- Context portability issue still limits value
- Stateful agents = $$$$$

https://t.co/EReYv5Vg5H

#ai #agents #llm #engineering

[Tom @Tom_MkV]

Trying out @jandotai for my local LLM server on my Mac Studio. I think this will be the single UI entry-point into my whole homelab...

[Tom @Tom_MkV]

Every second post is “introducing X” or “Breaking ����: Y”
Crazy times to be in the game!

[Tom @Tom_MkV]

I’m going to install https://t.co/aZQtGB1u2W in my 5 year old when it’s launched because he forgets everything, just like an agent….

[Tom @Tom_MkV]

Can't wait to show off what we're cooking at https://t.co/aZQtGB1u2W - the self-learning governable context layer for autonomous agent fleets. Each run makes your organisation a little bit smarter. Not a weekend project - but the solution to the context gap for greyfield dev.