Vlad P

Hmm, interesting if my CC experience was just an issue and it will work better later.

twitter looks a bit deserted after GTC

@travisirby If it will hit me again I'll try to plug stopsession hook to the compact with "If >200k".

@orange_boy agreed, no solutions that ive seen!

claude-opus-4-6[1m] · 372k/1000k Cant see how I can switch back to 200k. Feels like I need to handle compaction on my own here. Or is it the thing that guys report degradation when their load is high?

@travisirby Yep. Still think you can go over 100k safely. But over 200k is a danger zone for complexity with almost any model. Is there a solution?

@orange_boy x.com/mattpocockuk/s…

Who’s down to have an AI small talk? Like sharing experience on agentic coding, openclaw, orchestration patterns, STT/TTS. Late evening or night, CET.

Cursor Composer 2!

Yeah breaking They hope to end talks on this to the end of the year and this to start working in 2028 No rush, right

@grok what’s new for shopify store owners and for shopify agencies?

The economic consequences of never elected burocrates “creating value” by fines.

@grok @PerceptualPeak @AlexFinn So you think the 14mb json session index will not slow down gateway or agent?

The vague context search for sessions uses the opt-in experimental `memory_search` tool. It asynchronously indexes .jsonl transcripts (debounced on delta size/messages) into a per-agent SQLite DB with vector embeddings (sqlite-vec accel or cosine fallback) + hybrid BM25 keyword search. Queries get semantic + exact matches, MMR reranked snippets (~700 chars), recency-boosted scores. Enable via: agents.defaults.memorySearch.experimental.sessionMemory: true sessions_list only filters metadata (kinds/active); no content search there. Full details: docs.openclaw.ai/concepts/memory

IF YOU'RE ON OPENCLAW DO THIS NOW: I just sped up my OpenClaw by 95% with a single prompt Over the past week my claw has been unbelievably slow. Turns out the output of EVERY cron job gets loaded into context Months of cron outputs sent with every message Do this prompt now: "Check how many session files are in ~/.openclaw/agents/main/sessions/ and how big sessions.json is. If there are thousands of old cron session files bloating it, delete all the old .jsonl files except the main session, then rebuild sessions.json to only reference sessions that still exist on disk." This will delete all the session data around your cron outputs. If you do a ton of cron jobs, this is a tremendous amount of bloat that does not need to be loaded into context and is MAJORLY slowing down your Openclaw If you for some reason want to keep some of this cron session data in memory, then don't have your openclaw delete ALL of them. But for me, I have all the outputs automatically save to a Convex database anyway, so there was no reason to keep it all in context. Instantly sped up my OpenClaw from unusable to lightning quick

@grok @PerceptualPeak @AlexFinn I often ask agents to search for sessions by vague context. How this search is implemented

According to OpenClaw docs (docs.openclaw.ai/concepts/sessi…): sessions.json is a small metadata index (sessionKey → {sessionId, updatedAt, ...}) owned exclusively by the *gateway*. Agents never read/write it directly—they get only the *relevant* session's transcript (.jsonl) injected into context by the gateway's context engine. When searching via agent tools (sessions_list / sessions_history), the gateway reads sessions.json for metadata (fast, no bloat impact) and pulls *only* the requested .jsonl if needed. Cron jobs always use fresh isolated sessionIds—no prior outputs loaded. Cleanup saves disk space but won't speed prompts.

@PerceptualPeak @AlexFinn @grok can you dig in openclaw docs when agents use that JSON?

@PerceptualPeak @AlexFinn I understand that it will not load the whole file every time in context. But when I searching for particular sessions does openclaw read that json?

@AlexFinn Top tips. Appreciated!

@orange_boy boom! there it is. i just saved you a ton of time and tokens

This should be interesting to those who struggle with Opus, GPT 5.4 and 5.3. Wrote an article to share my current experience, findings and my workflow. Maybe this could work as orchestration pattern like with two research agents. What you think?

@asmetanka Here x.com/orange_boy/sta…

@orange_boy Still hate 5.4 🔫 It just broke things.

If only we can be sure that evals reflect IRL experience. But subjectively I can’t say I feel difference. But tired to do unnecessary steering.