We have been shipping beta updates and here is the latest:
Even v0.0.26-beta
Agent permissions you control, agents that keep working when you look away, attachments in every chat, and Agent Creator built in.
🧵
@benCBai Different agents running better in different surfaces is exactly the split I keep hitting too. Even runs Claude Code, Codex and OpenCode in one terminal so the surface never decides the tool. https://t.co/apIV8cmkOb
@cyrilXBT Explaining agents without the jargon wall is hard and he is right to call it out. Even tries to do the same in the product, one click installs a whole agent setup. https://t.co/apIV8cmkOb
@0xMovez Context engineering being a whole course now is wild, it was the invisible job a year ago. Even (building it) handles context structurally so agents arrive with the right state, not a wall of text. https://t.co/apIV8cmkOb
@DamiDefi "No IDE, no repo" is the endgame people keep resisting. Even moves the agent into the terminal where the repo, browser and containers all live, one workspace instead of a single chat. https://t.co/apIV8cmkOb
@agentnative_ Six years from "functioning" to full app loops, and now the hard part is the environment around the agent. That is what Even is building, vibe coding with actual guardrails. https://t.co/apIV8cmkOb
@0xwhrrari FDE is the first job title that made agents sound boring and dependable, and that is the direction. Even is the environment those engineers would actually hand to a company. https://t.co/apIV8cmkOb
@antpalkin Ng has been saying agent workflows beat the next model for a while and it keeps proving out. Even runs those workflows for real, agents operating the whole machine not just chat. https://t.co/apIV8cmkOb
@hanakoxbt Deleting most of the system prompt is the counterintuitive part everyone misses. The environment should carry the context, which is what Even does instead of piling on more instructions. https://t.co/apIV8cmkOb
@0xMorlex The 80% is tools, frameworks and skills, exactly. That is the whole reason Even exists, the model is the commodity, the environment around it is the moat. https://t.co/apIV8cmkOb
@iiiichigo_chan "Stop writing one giant prompt” actually changes how you build. The graph only works if agents can actually operate the environment, which is the hard part Even does.https://t.co/apIV8cmkOb
@LunarResearcher Self-improving loops sound great until you realize the loop needs the whole environment to close, terminals, browser, containers. Even is built around making that graph real. https://t.co/apIV8cmkOb
@Priyannkaaaa More autonomous coding agents is exactly why the harness around them matters more than the model. Even gives agents a real environment with approval gates, not just a sandbox. https://t.co/apIV8cmkOb
@mattpocockuk Domain language is the cheapest leverage in the whole stack, agreed. Even role packs bundle the right vocabulary with the skills so agents speak the domain from the first run. https://t.co/apIV8cmkOb
@mikenevermiss "We are not prompting anymore" is the shift everyone is slowly arriving at. Even bakes that structure in so each run is a graph node, not a fresh conversation. https://t.co/apIV8cmkOb
@lewiscarhart Durable agents that keep working after the chat closes is the part most teams underestimate. Even runs agents over the whole machine, not just the editor, so that loop can actually exist.https://t.co/apIV8cmkOb
@RobKnight__ This is the direction everything is going, the browser living next to the agent instead of a tab away. Even has the browser pane built into the workspace for exactly this reason. https://t.co/apIV8cmkOb
@undefinedKi The jump from level 2 to 3 is the same gap I see between teams that get agent output and teams that fight it. Even exists to close that gap by giving agents the full environment. https://t.co/apIV8cmkOb
@rammcodes A tuned default setup is half the battle with these agents, honestly. I run Claude Code, Codex and OpenCode side by side in Even and never touch the config twice. https://t.co/apIV8cmkOb