EdgeDimi

EdgeBuddy is the epitomy of sophisticated backend to "one click sync" to have access to a trading companion that connects the dots of your trading performance in ways that matter. Finally, someone who actually "gets" you 24/7.

@Liquid_Hwite @steipete @openclaw try the v5.3 works perfectly for me!

@steipete @EdgeDimi @openclaw yep, the may release is the first oc update that broke my oc and had me edit the config manually and warped me back in time early claw days

I don’t know what’s going on with the last week of @openclaw releases. The feature direction is genuinely exciting, but the base keeps shifting so fast that every release I tested hit a different foundational break. - v2026.4.27: local memory/search broke because the managed runtime deps did not retain/resolve node-llama-cpp. -v2026.4.29: that class looked fixed, but Discord/Telegram failed on startup because packaged plugin runtime deps missed json5. Repair attempts were not durable. -v2026.5.2: channels could be made healthy only after local patching, but native Codex/tool adoption hit a worse blocker: tools visible in catalog/runtime inspect were missing from tools.effective, so agents could not reliably use memory/wiki tools. These are not edge-case polish issues. They are “can the agent actually operate normally after upgrade?” issues. Right now testing the new features feels impossible. Stability first, then features. Back to v2026.4.26

@openclaw introduces its first "airdrop" style plug in for trusted connected paired nodes for file transfers. Brick by brick! Additionally all obstacles holding me back to v2026.4.2026 are all resolved time to test google meets plugin and more on a real stable version 2026.5.3! A small hiccup if image_generate succeeds but Discord/Telegram attachment delivery fails with: "Optional dependency sharp is required for image attachment processing" , check whether sharp resolves from your active OpenClaw package root, not only the plugin npm root. Thats an easy hot fix. Finally my first Codex harness setup! { "agentRuntime": { "id": "codex" }, "model": { "primary": "openai/gpt-5.5" }, "plugins": { "entries": { "codex": { "config": { "codexDynamicToolsProfile": "openclaw-compat" } } } }, "messages": { "visibleReplies": "automatic", "groupChat": { "visibleReplies": "automatic" } } } I did not start with native-first tooling because my agents already rely on OpenClaw tools like memory, workspace read/edit, message delivery, browser, and exec-guardian policy(custom made plug-in). Native-first is interesting, but openclaw-compat is the safer bridge for testing the Codex harness without removing the existing OpenClaw operating surface. happy days ahead! thanks @steipete !!

@Voxyz_ai Thats good ❤️

@EdgeDimi I checked the actual agent trajectories from yesterday’s 5.2 run and today’s 5.3 run. Both loaded the same 10 effective tools: read, write, edit, exec, web_fetch, sessions_list, sessions_history, sessions_send, memory_get, memory_search. it didn’t repro on my setup.

2026.5.3 dropped. big release, fewer paper cuts. a few features worth flagging, plus my fresh bench numbers below: → /steer: redirect mid-session without restarting the turn → /side: ask side questions in the same context, no main transcript pollution → file_fetch / file_write: move files across paired nodes → plugin install/update hardened: official npm + ClawHub fallback + beta channel + doctor --fix → usage / session views got lighter, cost now from persistent aggregate cache vs 5.2 where three CLI → Gateway rounds all landed 10-13s, today's 5.3 runs: 10.8s / 9.8s / 10.4s average 10.3s vs 11.6s. model itself tightened from 2.9-6.6s to 2.3-3.7s, more stable session lane startup still has that fixed 6-7s wait. from the trajectory it reads like Gateway queueing the agent run, not the model that said, telegram on my end feels worse this round

@Voxyz_ai nah it was open issue after v5.2, tools effective was not surfacing the in built tools memory_search/ memory_get and wiki tools so if you had to surface sth from your memory layer it was not possible

@EdgeDimi 5.2 ran fine on my end, 10 tools loaded both runs. yours sounds config specific?

@fagamericano @steipete @openclaw Good old monolithic 😂

@steipete @EdgeDimi @openclaw I have our enterprise version still pinned on 4.23 until the dust settles :(

@Strong_Genetics @steipete @openclaw Depends on your configs and how much end to end testing you did. After every release i go run every possible scenario as smoke test. On v5.2 memory_search tool was the blocker to roll back , maybe if you also ask your lobster to retrieve sth you will hit the blocker ;)

@steipete @EdgeDimi @openclaw That's so weird. 5.2 has been great for me. I've been having problems ever since 4.24+. I don't know how you guys deal with all these variables but keep up the good work. We all know it's hard

@steipete @openclaw How approachable you are to people that really embrace your creation and we all collectively try to bring it to its best form... just no words Peter! Grateful!

@EdgeDimi @openclaw That's on me. I started hiring people to professionalize this and help with more hands. The May release is the first one where you can feel the difference.

@BenjaminBadejo @mim757s broken, majority of in built tools not found under effective tools so the agent fails to do a simple memory_search. first time since January i have done 3 roll backs v4.27 v4.29 v5.2 all broken releases

@mim757s Try 5.2

My little macOS apps for switching OpenClaw from the Pi harness to the Codex harness on the fly, including switching the agents.defaults.models block so I don’t have to touch the config file at all. Just one click and then /new in any agent chat. Convenient.

The official repair path does not produce a durable fix. It can appear repaired briefly, but after a Core restart the runtime deps get rebuilt incorrectly again, json5 disappears from the effective runtime tree, and Discord/Telegram crash. There are open PRS so this will be fixed soon I guess

@elvissun @openclaw Browser_use tool of openclaw with custom browser mastery skill teaching the agent all browser actions and using separate authenticated openclaw profile with your LinkedIn credentials. It uses the account like a person. Have noticed way quicker turnarounds with gpt 5.5 vs opus.

@EdgeDimi @openclaw still experimenting, might need to run some evals on exa. hunter and apollo are too expensive probably. computer use is another option but i don't want to burn my account

my @openclaw-powered CRM is now multi-channel. every new signup now gets a linkedin lookup. brave search first, codex agents in parallel as fallback. profiles delivered in my morning telegram. this might be the highest ROI growth hack I've done this year. warmer leads, real connections, linkedin growth, all for $0/mo. funny that we can now use agent swarm to play the linkedin game

GM ! Not opening duplicate PRs for no reason but this one is operationally critical : -daily session reset must survive heartbeat/system events. Running WhatsApp + Discord agents 24/7, if updatedAt gets touched after 04:00 while the JSONL started yesterday, sessions never roll. Verified this on 4.23 and carry a local hotfix until upstream fixes #65386/#68315. My local hotfix stamps sessionStartedAt from the JSONL header, makes daily reset compare against min(sessionStartedAt, updatedAt), and removes the system-event shortcut so heartbeat/cron can’t mark an old session fresh before rollover.

Built clawsweeper, which runs 50 codex in parallel around the clock, scans issues/prs deep and closes what is already implemented or what makes no sense. Closed around 4000 issues today, a few thousand are in the pipeline. (rate limits are rough) github.com/openclaw/claws…

Private local set ups on open source software. Thats the agent architecture. Dont fall for ease of use especially if you own a company privacy and security oriented.

Infra agnostic model agnostic is he way. Own your stack

Have an architecture that no vendor can stop your harness. Thats the craft

@Andrey__HQ Definitely greek with the variation of Greek salad there. Enjoy!

I hate meal prepping so I made a deal with my friend Helena I paid her $50 to come over and meal prep for me while I was clauding and look at this beauty I’m now set for the entire week Claude could never do this

@Andrey__HQ You need a parallel browser running for agents, now anything headless gets bot detection. And since you are way deeper inti this do we need a new briwser architecture or rewrite of all the websites and how they used to treat "visitors" as humans?

Hot take: the future internet for agents won't be visual. the problem I keep seeing with how people think about web agents is that everyone is trying to make the browser faster for machines. there's at least 100 threads on terminal browsers, headless rendering, etc.. some might say that it makes sense since the browser is how we interact with the web so it feels natural to make agents use it too. but agents don't have eyes. in other words: they don't need a visual interface at all. e.g: when you book a flight on jetblue, your browser sends a POST request to their API with a JSON body containing your origin, destination, and date. The server sends back structured data with prices, times, and availability. Then your browser takes that structured data and renders it into a visual layout with buttons and colors and fonts so that you, a human, can read it. pretty simple. but an agent doesn't need that rendering step. all it needs is the structured data that was already there before the browser painted it on screen. The entire visual layer is a translation step that only exists because humans need it (because of trust) this means the parallel internet, the one where agents are transacting on behalf of people, won't look like a faster version of the human internet. It won't look like anything at all.

So companies won't pay 5k for an implementation agency to put with experience custom workflows in a week, they will pay 5-10k for an employee to get trained for 2 months amd then wait this employee to have gained the same experience with a specialist to apply the workflows to the team. How you think these economics make sense?

@JohnWGant you could certify 50 people in a 8 week cohort style program for $5-10k/person or you could do one custom build for $50-100k that takes you 50 hours and a full team pretty easy choice

the real play here isn't building custom AI for businesses. it's certifying the people inside the business to do it themselves. skip the scope creep. skip being on call when something breaks. just train their ops person, their EA, their marketing lead to implement AI internally. less revenue upfront. infinitely more scalable. and the business actually retains the knowledge instead of depending on you forever. someone is going to make a killing with this model.