contexto 🦞

Today we are releasing Contexto v0.1.11 Contexto is an OpenClaw plug-in with smart context management. With Contexto, keep long-running OpenClaw agents reliable even after the context window fills. Sign up now. Link in the comments!

Recently, we hosted our 7th OpenClaw Demo Night, this time in Shanghai. 🦞🇨🇳 Our biggest one yet. 120+ people joined us, and 8 builders from different parts of the world demoed the projects they are building with OpenClaw. Hosted at the Alibaba HQ in Shanghai. Nothing feels better than creating a platform where people showcase cutting-edge work, exchange ideas, and push the entire ecosystem forward together. This is exactly why we started doing these events.

today, i presented what we are building at @getcontexto and shared my journey of hosting openclaw events around the world. a room full of tsinghua and peking students, researchers from leading chinese labs (z ai and deepseek), and builders at the epicenter of china’s ai and academic scene. so grateful to vibe friends for organising this event and inviting me.

1/21 @harshilanand39 harshil used openclaw not just to automate his own workflows, but to serve his clients' needs too. for a crypto recruiter, his agents auto-prepped every candidate interview and sat inside 400+ telegram chats, catching hiring signals the founder would've missed. for his own agency, a team of agents is building v2 of the company from scratch.

at lobsterpod.net we hosted openclaw demo nights and meetups across 6 countries.🦞 today we start a series where we show the top 21 openclaw demos from our events.

What counts toward the context window Everything the model receives counts, including: -System prompt (all sections). -Conversation history. -Tool calls + tool results. -Attachments/transcripts (images/audio/files). -Compaction summaries and pruning artifacts. -Provider “wrappers” or hidden headers (not visible, still counted).

Context is not the same thing as “memory”: memory can be stored on disk and reloaded later; context is what’s inside the model’s current window.

@PrivanaFinance privacy is freedom

1/ Your private corner of DeFi Get early access

The result: constraints survive compaction. Tool outputs stop bloating the window. The agent stops repeating work it already did. Context doesn't collapse as the session grows — it compounds. We're building this at @getcontexto Sign up for early access → getcontexto.com/signup

How it works: Stores complete agent turns as "episodes" (not individual messages) Keeps a small sliding window of recent work Moves older episodes into Mindmap, an indexed long-range store using hierarchical clustering Auto-retrieves the right past episodes based on relevance to the current step Nothing is flattened into a lossy summary.

Your AI agent gets worse the longer you use it. Not because the model is bad. Because of context collapse. Old constraints get buried. Tool outputs pile up. Decisions from step 3 get contradicted at step 7. The context window has a hard limit but degradation starts long before it's reached.🧵

the email agent had one rule: don't delete without approval. tool outputs piled up. the rule got buried. then dropped during compaction. agent bulk-deleted hundreds of emails. it did exactly what the context told it to. the context just stopped telling it the right thing.

you could work with the smartest person alive. give them the wrong brief and you still get a bad answer. that's what's happening with AI agents right now. the model is fine. the input is the problem.

your agent didn't hallucinate. it didn't rebel. it just couldn't see the constraint anymore. that's context collapse.

models keep getting better and more interchangeable. context doesn't. context compounds. own your context and you stop being a consumer of AI. you start building on top of it.

context compounds. every task your agent runs generates signal. what was needed, what worked, what didn't. a context engine captures all of it. over time it builds up your domain knowledge, your workflows, your decision patterns. you stop being a consumer of AI. you start building on top of it. that's what we're building at @getcontexto getcontexto.com/signup

real example. email agent. user says "don't take action without my approval." agent starts processing. searching, categorizing, flagging. tool outputs pile up. the original constraint gets buried then dropped during compaction. agent bulk-deletes hundreds of emails. it didn't hallucinate. it didn't rebel. it just couldn't see the rule anymore.

your AI agent isn't broken. you're just feeding it bad context. you could work with the smartest person alive. give them the wrong brief and you still get a bad answer. that's what's happening with agents right now. the model is fine. the input is the problem.