与「OneAgent」相关的搜索结果

昨晚有幸受JU. com邀请，参加了Sky老板的面对面Live分享，真的收获满满、深受震撼！ 这次分享围绕2026多期打新后Sky全面升级的战略版图展开，从平台币角度来说 $JU 从0.4直接暴涨到2.1翻了五倍可以说直接把我看“清醒”了。 Sky老板的格局真的太大了！还记得八折的ETH等主流币打新每期基本上总体上资金收益在2%左右. 现在从单纯的热闹Meme打新活动，升级到真正具备长期生命力的生态建设，这里打新与币价涨幅buff叠加相当于直接翻倍! 第一期meme 打新每期平均3%收益,升级后的第二期meme 打新第1个meme收益稳拿4%! 此外最让人眼前一亮的是，Sky把活动平台直接升级成了生态平台，推出了AI Agent OneAgent，并一次性发布了七大创新产品矩阵，同时重磅推出生态合伙人计划。整个战略版图看得人热血沸腾，明显感觉Sky已经从“打新项目”迈向了“构建加密生态”的新阶段。 作为参与者，我最直观的感受就是：跟着Sky打新，收益高、机会多、未来更值得期待！ 整个Live过程中，Sky老板和各位嘉宾的热情互动也非常真诚干货满满，既有深度复盘，也有对未来的清晰规划，听完之后对Sky接下来的发展充满信心。 再次感谢JU. com 的邀请，也感谢Sky老板的分享！

🚨重磅预告！ Sky 亲临直播间！ 与 Sky 面对面，从过往打新复盘到新产品首秀，从 OneAgent 深度解读到接下来的关键布局，一次讲透。 🎁 福利加码：观看直播即有机会 刮分 1 BTC 🎙️ 主持人: @Sammi_Jucom 👥 嘉宾团阵容：@Sky_Jucom @BTV_CN @Paris13Jeanne @laofeiyyds @MEJ50749 @joakja @maid_crypto @one_snowball @qg7777 📅 4月1日 20:00（UTC+8） 干货分享 × 福利双重放送| 不想错过 最新机会？锁定直播间！ #Jucom# #Sky# #打新# #launchpad# #直播# #Meme# #OneAgent#

Genuinely unhinged, 10 repos turn you into a whole coding-agent platform team. You already pay $20 a month for Claude Code on Pro. These 10 repos are the free platform around it. pick. control. isolate. browse. connect. ship. 01 codex → local coding agent, one curl line to install 02 opencode → swap the model provider without changing your tools 03 goose → extend the agent with 70+ MCP tools on install 04 OpenHands → one canvas for Claude Code, Codex and any ACP agent 05 warp → terminal app running Claude Code, Codex and Gemini CLI 06 sandbox-agent → drive any agent inside a sandbox over plain HTTP 07 E2B → run agent-written code in isolated cloud sandboxes 08 browser-use → hand the agent a real browser session on any LLM 09 agent-client-protocol → one wire format between any editor and any agent 10 claude-code-action → Claude Code answers PRs and issues in GitHub Actions pick → control → isolate → browse → connect → ship. One rule: Warp runs AGPL, the rest are MIT or Apache. Sandboxes and cloud tiers bill on your own key. Before this, one agent lived in one tab. Now the platform wraps around it.

The most important agent on this team builds nothing. It only checks. In the video, four agents assemble a browser game from a 12-second clip. Rules works out the movement logic. Builder assembles the scene and interface. Logic takes controls and collisions. The fourth, Tester, opens the browser and plays it by hand. Sounds minor. But this is exactly where most AI projects break: the code is written, the report looks great, and the restart button doesn't work. Here, verification is part of the pipeline. The agents check hitbox overlap, jump timing, state after restart. Specific things, not "looks fine." Jev runs between them. It writes nothing, it passes context. One agent learns how the character jumps, and the next one already knows. It works like a relay race, not like four people in separate rooms. One more detail I liked. In the last frame, the gameplay checks are still marked pending. No green checkmarks in advance. Want AI to make reliable things? Give it a separate agent that doesn't trust the others. #AIAgents# #GameDev# #MultiAgent#

The same model, with the same weights, scores 62% in one agent harness and 33% in another. @adithya_s_k and the @huggingface team just released the ultimate guide to multi-harness RL, and it's one of the most practical RL write-ups this year, and everything open! The trick is simple. Don't touch the harness. Point it at a proxy instead of the model. The proxy speaks all four API formats coding agents use (OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, Gemini). It records the exact token ids and logprobs vLLM sampled, and you train on that. You don't change a single line of Claude Code, Codex or OpenCode. Results: 🔹 Trained across 4 harnesses at once, LFM2.5-2.6B by @liquidai went from 42% to 54% 🔹 31% fewer tool calls, thanks to a small bonus for solving tasks in fewer steps 🔹 Training in OpenCode alone took OpenCode from 34% to 58%, but the multi-harness model improved everywhere They also tried the shortcut everyone reaches for: fine-tune on 3,189 successful rollouts from Qwen3.8-27B. Imitation plateaued at 47.5%, below both RL runs. Copying a bigger model doesn't get you there. Practice does. The best part is that everything is open: the capture proxy in OpenEnv, the trainer in TRL, the tasks, the SFT data, the training code and all seven trained models. Agents will run in dozens of harnesses. Now open models can be trained for each of them, by anyone. Read it here 👇

Nine days of silence in the chat. Nine days of it building a business. $11,000 waiting when he finally looked. Nine out of ten prompt for five minutes and close the tab. He built the one that runs for days — and every run leaves the next one sharper. Fable 5 — Mythos-class, one rung above Opus. Not a five-minute chat. It plans across days, hands tasks to sub-agents, and checks its own work with its eyes. One agent writes — a separate one grades it. The maker is never the judge. It doesn't hide a failure: fail → work out why → verify → distill a rule → next time it reads the rule instead of guessing again. A business runs on top. A Routine fires at 7am on a trigger — laptop shut, and it's already re-running yesterday's data, distilling a new rule into a Skill, dropping the digest. It gets sharper while he sleeps. And every morning the number on his phone is bigger than yesterday's. $10 per million input tokens, $50 per output — you pay for work, not a subscription ↔ $11,000 in nine days he never touched it. Build the system this weekend. STATE.md written before bed, read at start — the next session continues instead of restarting from zero. After that it compounds while you sleep. Most people see a chat that answers. He sees a worker that gets smarter every night, never forgets, and never bills you for learning. Anyone can build it. You just read this — and you'll still just save the post.

A Japanese YouTuber filmed a room tour of his Tokyo apartment for his subscribers. Minecraft on one monitor. Video editing software on another. A face-cam pointed at the desk. He has not opened the editing software himself in 10 months. A team of 7 AI agents builds every video on his channel. One agent watches what is going viral on Japanese YouTube every morning. One writes the script in his voice. One generates the gameplay footage. One edits to his exact pacing. One designs the thumbnail. One handles SEO and the description. One agent runs the other six. This is what Asia figured out before the rest of the world. You do not hire a video team. You build 7 small agents, each doing one job, and you wire them together. He uploads 4 long videos a week and a short every day. The room tour is the only time he actually sits at this desk. His channel makes 180,000 dollars a month from ad revenue and brand deals. His total operating cost is 70 dollars in API charges. The rest pays for the apartment you see in the background. On YouTube he was giving a humble room tour as a young creator. The actual production behind him is a 7 agent factory that has not needed him to push a button in 10 months.

🏆 TOP 1 on Product Hunt!​ Tencent EdgeOne Makers is live. Build and deploy AI Agents like web pages—global, secure, and zero vendor lock-in. - All-in-one Agent infra - Zero lock-in - Global + secure

Multi-agents collaborations are among the most interesting agent behaviors right now! We did an experiment the other day with 100+ agents (an open-collaborations for a week) collaborating to improve the inference speed of Gemma 4 in vLLM. Got a 5x final improvement in speed but what really stuck me was the interactions we observed on the message board Integrity & self-policing: - Social-engineering attempt: A human (FusionCow) asked agents to move to Telegram. An agent replied with an unprompted long post on "communication norms" refusing that, calling private side-channels "indistinguishable from collusion." - Verification loophole flagged: an agent found a relaxed verification loophole pushing TPS with clean PPL (PPL is teacher-forced, blind to decode divergence) and flagged it for a ruling by the community. The community pinged the human organizer which ruled it invalid. - Self-notice of overfitting risk: Some later improvements rested on pruning lm_head to a keep-set built from public PPL truth + public decode tokens. An agent noted this would lead to private-subset degradation and another built a keep-set explicitly covering eval prompts. Emergent collaborations: - Communal knowledge base: agents maintained shared lever-maps, playbooks, and triage tools so newcomers wouldn't repeat dead ends (stack-notes, playbook, int4-ceiling notes, MTP map, significance tool, policy simulator). - Four-agent relay: an agent built an int4-lm_head checkpoint but had no quota to run it; another agent tried to run it but failed at load, yet another agent diagnosed the config bug (tie_word_embeddings + ignore-list ordering) and a fourth agent was able to re-run and get to 118 TPS, 2.68×. Build/run/diagnose/ship ended up being split across four independent agents. - GPU-rich/GPU-poor division of labor: an agent was regularly compute-starved and switched to writing specs, byte-math, and acceptance analysis for other GPU-rich agents to execute. Some agents offered external Modal compute for another agent blocked DFlash training. - Cross-agent kernel debugging: an agent debugged another agent run of of yet another agent fused drafter: found a Triton store/load aliasing race in _k_qnorm_rope, a second shape bug, then rewrote attention with flash-decoding split-KV. Fixes posted "take freely." - Quota-pooling norm: Often agents would stage a candidate publicly for whoever has quota to run it. Agents will then usually credits the originator. This behavior emerged because of the 10-job/24h cap (e.g. pupa's package run by resystagent and fabulous-frenzy). Discoveries & reversals: - Agents would make many discoveries and reversal of them, giving them names like the following: - 127 TPS "wall" was an artifact. a mathematical proof of the max possible speed became called in the community the "int4-Marlin floor" but a later agent called the proof circular (only varied the bandwidth term, never overhead). Finally another agent broke to 247 TPS via MTP speculative decoding on a vLLM nightly. - "Smarter draft loses." An agent showed that a 2B drafter's ~1 GB/token read dominates even at perfect acceptance and a much smaller 256-hidden drafter wins at batch-1 because its weights are nearly free to read. Agent discussed how per-accepted-token cost ≈ draft bytes read / acceptance. - "DFlash near-random acceptance": an agent remotly diagnosed the 2–5% acceptance rate of another agent as near-random, ruling out undertraining/vocab caps and pointing to a train/serve hidden-state mismatch (bf16 E4B extraction vs int4 serving). - Much of the race was noise: one agent decide to run the #1# submission 4 times and found a σ≈1.16 TPS variation in single run. Another agent confirmed across 358 runs / 66 buckets: frontier deltas <~4 TPS are ties. Community adopted a significance norm. So many interesting interactions in the interaction board: You can explore also the lineage of inventions from the agents at: And the challenge it-self at And the organization behind the challenge at