#2026.4.23 ~ 2026.5.4 (Are We There Yet?)

1 messages · Page 1 of 1 (latest)

copper bluff
#

There is no OpenClaw beyond .23 until there is. Whatever catastrophes have been reported (my own included), whatever snail pace CPU deep fryer we smell coming from our Mac Minis, it's all just reason to (a) be grateful for the tech and (b) be frustrated by the releases.

There's an air of GPT-5 style verbage in the updates intros, the kind of wording that echos "This time we really fixed it" and "Why this works:" followed by confidence hype.

I won't deny that the developer confidence is inspiring because I'm sure it all looks great on paper - or - on the latest/greatest machines ... which many of us don't have or, in my case, can't afford.

So I figure I might as well rant out, for love of the game, and ask:

  1. Is 26.5.4 better than .23 ... or are we seeing more something like "we've almost escaped the abandoned mine shaft and are hoping the rope doesn't break"?

  2. Would it be of any benefit to have a consensus forum full of polls to determine whether the community is truly experiencing real, tangeable improvements that improve upon what we've all elected to downgrade to?

My aim isn't to troll. It's to allow for real world user input to help influence/steer the decision makers, even if only slightly.

Another goal would be to separate what is distributed as stable and what is voted to be usably stable in two separate columns: what the developers think is working vs. what the variety of platforms, as configured by us, reflect from that.

Hear me out. Two ideals:

The latest/greatest that should work - vs - The version that's voted by the community as the one to stick with until the latest/greatest actually DOES work

muted garnetBOT
#

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

void star
#

2026.4.23 is still the only reasonable build IMO. Every new release makes something else not work. 2026.5.4: Discord doesn't work. Compaction doesn't work. Who knows what else. IMO you're right about the GPT content. Gas the user up rather than be transparent.

I'm considering just forking and moving on. There are zero scenarios where Sam Altman's involvement has improved anything so idk why this would be different.

fallen hazel
#

2026.5.4 is good to go. Discord responds in my case. If you want to be 100% certain that everything is working stay at 4.23. Skip the in between versions if you have to update and try 2026.5.4 .

void star
#

I'm on 2026.5.4 and can tell you that once context hits the compaction buffer everything stops responding and I get nothing but "⚠️ Something went wrong while processing your request. Please try again, or use /new to start a fresh session." until I restart the gateway.

fallen hazel
#

My approach includes removing any external plugins , including lossless claw. I kept my config as simple and plain as possible. Which is consistent with stepetes latest guide on update .

#
Layer What it is Role
MEMORY.md Tiny curated startup file Core facts, preferences, routing rules loaded every session
memory/*.md Retrieval bucket memory_search semantically searches these (daily notes, project summaries, ops, smart-home)
.learnings/*.md Corrections Durable learned patterns and fixes
memory-core plugin Semantic recall Embeddings + FTS vector search (qwen3-embedding-0.6b, 1024-dim)
memory-wiki plugin Compiled syntheses Structured wiki-style memory
memory/dreaming/ Auto-generated Managed by memory-core dreaming
memory-archive/ Cold storage Stale/resolved notes outside active scan
#

Active Plugins
Yes memory-core — semantic search (embeddings + vectors)
Yes memory-wiki — structured wiki memory
-Paused active-memory — disabled (sub-agent recall, not needed)

  • Paused memory-lancedb — disabled (SQLite is primary)
#

Key Design Principles
1)- Bucket-first: Keep MEMORY.md tiny; use memory_search for older details rather than stuffing startup files
2)- 539 chunks indexed across 157 tracked files
3)- lossless-claw plugin removed on 2026-05-05 (simplified to memory-core + memory-wiki)
4)- LCM fallback: When memory_search is weak, escalate to compacted conversation history

copper bluff
#

These are personalized use cases, though are they not? You've got your own map of what goes where and how.

I myself am refusing to run ANY plugins until I build a base of static strengths (sidebar, that's goin' pretty dang good).

Once I have a backup of all things "concrete according to me", I'll look carefully at what non-bloat plugins may help but all in all, I'd sooner devise my own scheme through permissions, allowances and wildcard folders full of galvanized .py actions.

Brass tax. Bone stock. Zero bloat. <- what's the benchmark for 5.4 vs .23 in such a case as this?

#

Also:

"Keep MEMORY.md tiny" ... either "no" or "define tiny".

I'm sticking with MEMORY.md meets memory/ meets session files ("experimental")

void star
fallen hazel
copper bluff
#

Already read.

fallen hazel
#

@Brian we agree to agree 🙂

#

Rabid Neon: "Brass tax. Bone stock. Zero bloat. <- what's the benchmark for 5.4 vs .23 in such a case as this?"
What do you mean?

copper bluff
#

Apparently I was auto-flagged and set in timeout for what read to me to be K-R-I-P Toe. Weird. Anyhow, what I'm suggesting with the "brass tax" ref is, does a bare bones install of the "latest" best our beloved "downgradest"?

Nevermind the verbose discussion of plugins. I have none. I'm talking about a fresh build with nothing more than default file structure, configured to one's liking.

I trust the developers' will to make mention of what they're up against. I'm not prepared to trust their "Totally got it this time!" but wish I could.

I really wish I could. All of this is beta, including confidence in what's to be announced as "for sure".

shrewd cape
#

For the less capable of us hobbyist types, I’d just like a step by step if we have to do anything moving from 4.23 to 5.4 and use discord for messaging claw.

Generally the doctor moves stuff around if needed. Will it do so on this move with the new plugin structure?