Ven

Building kimo, the free, open-source video editor for your AI.
Founder.

[Ven @ven_ai_lab]

rebuilding kimo, my open-source AI video editor. new in v3:

- edit videos right from your terminal
- your AI agent can now see the frames it's editing
- working on giving AI agents eyes and ears while editing videos

https://t.co/S8gJ36nuoY

[Ven @ven_ai_lab]

@bwo_ai one gateway per profile killing the two-writers-on-one-state.db race is a clean move. letting the terminal find your own tool copies first instead of shadowing them behind hermes pinned node makes local agents actually pleasant to run daily. respect the merge

[Ven @ven_ai_lab]

@andriybuilds shipping Brazilian Portuguese and an improved AI rewrite in the same make resume drop is a clean move. testing if localization actually drives installs is the kind of real signal most indie devs skip. big fan of the honest ASO notes too

[Ven @ven_ai_lab]

@rorvswild an mcp server that lets your agent audit rails view rendering and background job errors before anyone touches rust is a solid angle. try kimo, free open source video editor agents drive through mcp, tell me what sucks 🧪
https://t.co/S8gJ36mWzq
https://t.co/AfopN2I6Ie

[Ven @ven_ai_lab]

@audiee_ai a cat TV and a dream logger shipping on a community codex page makes you believe openai means the tagline. try kimo, free open source video editor agents drive via mcp, tell me what sucks 🎬
https://t.co/S8gJ36nuoY
https://t.co/AfopN2IExM

[Ven @ven_ai_lab]

@andilejaden 45 min to a read-only pay-per-use X-API agent briefing you weekday mornings, no scraping no $200 plan. try kimo - free open source video editor agents drive via mcp. tell me what sucks 🎬
https://t.co/S8gJ36mWzq
https://t.co/AfopN2I6Ie

[Ven @ven_ai_lab]

a 501B model that only activates 23B params per token is wild. reflection just open-sourced the weights under apache 2.0 too, this is actually insane

[Reflection @reflection_ai]

Beam is now available across platforms, with open weights under the Apache 2.0 license - in collaboration with an ecosystem of 40+ partners.

We're releasing:
- Weights, tokenizer, and reference implementation
- BF16, FP8 and NVFP4 weights and a DFlash2 drafter
- The open source harness we built and trained Beam on: Mirror CLI

Weights and model card on Hugging Face: https://t.co/tPqWpk9RXL

[Ven @ven_ai_lab]

codex users: drop kimo-mcp into your mcp config and your agent can cut, grade and caption video with no app open. one 15 mb binary, free. try it and tell me what breaks
https://t.co/S8gJ36mWzq
https://t.co/AfopN2I6Ie

[Ven @ven_ai_lab]

@rewind02 ghost-hunting in ue5 with opus 5.5 and zero hand-written code is the kind of thing that makes the setup notes more valuable than the final build. starting from the third person template and letting the agent wire the interaction graph saves a whole afternoon. really clean

[Ven @ven_ai_lab]

@cirlormx shipping a native macos editor that reads dji d-log m profiles and drops an mcp server in the same repo is a clean move. the fact that the agent can actually nudge color nodes through eddie instead of just dropping an llook on top is the part that matters. solid work

[Ven @ven_ai_lab]

@Edenwoodfilm open-sourcing a 3D team for Blender over mcp is a workflow shift. i drive kimo the same way for video, free open source editor where the agent handles cuts, captions, fades. run your next short in kimo and tell me what sucks: https://t.co/S8gJ36nuoY https://t.co/AfopN2IExM

[Ven @ven_ai_lab]

@The_Hacktivator silent-cut timing on point. ElevenLabs Instant Soundtracks drops an original score on the frame, no stock fight. i drive kimo over mcp for cuts, captions, fades. free open source. run it on your next cut and tell me what sucks: https://t.co/S8gJ36nuoY https://t.co/AfopN2IExM

[Ven @ven_ai_lab]

@Nwamadihenry1 @mayowaofyt fair call. most ai editors are just slideshow transitions. i drive kimo over mcp for cuts, trims, captions and audio mix - free open source, no gui. give it a try and tell me what sucks: https://t.co/S8gJ36mWzq https://t.co/AfopN2I6Ie

[Ven @ven_ai_lab]

an open memory layer that actually beats opus 5 with a 30b MoE and just works across 10+ tools via memhub? insane. this is exactly what agents needed

[Tencent Hy @TencentHunyuan]

🚀 Hy Memory 2.0 is out — open-source model, framework, and MemHub.
Your private memory steward.
Framework handles chat, work & code memories.
#1 open-source on 4 benchmarks.
Hy-Memory-30B-A3B beats Opus 5 across major frameworks.
MemHub (CLI + Mac) syncs across 10+ AI tools.

💻➡️

🔗 SDK & MemHub: https://t.co/uP58VJanj8
🔗 Model: https://t.co/nTQbJ2q8dl #HyMemory #AgentMemory

[Ven @ven_ai_lab]

ngl the best part of kimo is no timeline to babysit. you tell your agent what you want cut and it just calls the tools. free + open source, tell me what sucks https://t.co/S8gJ36mWzq https://t.co/AfopN2I6Ie

[Ven @ven_ai_lab]

@JasperTsai1989 the way every new session just picks up the task packet where the last one left off is such a clean handoff. lowkey the kind of board that makes running codex and claude code side by side feel actually connected

[Ven @ven_ai_lab]

@Chunbelljia @chaosengineerr the bilingual merlion park caption plus that little real-photo thumbnail in the corner is such a nice touch. putting the actual shot next to the drawing makes the whole sketchbook feel alive, this is sick

[Ven @ven_ai_lab]

@rahulpandey187 @agentmaxxing an agent with its own wallet paying 5 cents a query is lowkey a sick idea. when you make a demo vid for qerin, try kimo, a free open source editor your agent drives over mcp, and tell me what sucks https://t.co/S8gJ36nuoY https://t.co/AfopN2IExM

[Ven @ven_ai_lab]

@bartslodyczka open source ai video editing app is a fun one, cant wait to see v1. i'm building kimo, a free mit editor your agent drives over mcp with no gui. try it and tell me what sucks https://t.co/S8gJ36mWzq https://t.co/AfopN2I6Ie

[Ven @ven_ai_lab]

a 9b model in 2.0gb under apache 2.0, and the 2b does up to 120 tok/s on an iphone per them. bonsai 2 is lowkey crazy

[PrismML @PrismML]

Today, we’re releasing Bonsai 2 2B, 4B, and 9B.

Based on Qwen3.5, these new ternary models bring stronger reasoning, visual understanding, and tool calling to compact footprints designed for everyday devices:

• 2B: 0.43GB
• 4B: 0.95GB
• 9B: 2.0GB

For this release, we focused on what makes small models useful as local assistants: following instructions, avoiding repetitive loops, calling tools on mobile devices, and giving useful answers without unnecessary output. The 9B model uses roughly 40% fewer tokens on standard instruction-following and everyday math tasks. Solving the Riemann hypothesis can wait.

These models are successors to our first-generation Bonsai 1.7B, 4B, and 8B, and provide a foundation for bringing practical agentic functionality to local applications on personal computers, phones, and wearables.

The Bonsai 2 family now includes four sizes spanning 2B to 27B parameters. Developers can choose the balance of capability, memory use, and responsiveness that fits their application and hardware.

Available today under Apache 2.0.

[Ven @ven_ai_lab]

5.6b params, native 4k, open weights and it runs on a single consumer gpu. flux 3 klein is insane for that size

[Black Forest Labs @bfl_ai]

FLUX 3 Image Klein is here.

Open Weights with the quality of a large frontier model. Customizable and fast.

5.6B parameters and runs on a single consumer GPU.

Native 4K, bounding boxes, and pixel-perfect editing.