> uv run mklang console --provider deepseek --workspace /tmp/mklang-agent-machines
 ⭘                                         mklang console
 READY · deepseek · tokens 0+0 · session 20260718-013216-8c6d
 Ready. Ask the agent what should happen, or type /help.
┌────────────────────────────────────────────────────────────────────────────────────────────────────┐
▊  Ask the agent or type / for commands                                                              ▎
▊▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▎
 f2 Inspector  ^t Activity  ^g Stop  ^l Clear                                              ▏^p palette
▊  Find this week's news about open-source AI models and give me a short brief with links.           ▎
 RUNNING · deepseek · tokens 0+0 · session 20260718-013216-8c6d
 you: Find this week's news about open-source AI models and give me a short brief with links.
 ▼ Find this week's news about open-source AI models and give m
 └── ▼ console_agent
 RUNNING · deepseek · tokens 1630+575 · session 20260718-013216-8c6d
 ⏸ Consent: machine 'news_search' will call host tool(s) [search] (e.g. live web search). Allow
 this once for the session?  → type y / yes / sì  (Enter = no)
     ├── ● decide ok → discover
     ├── ● discover ok → decide
     ├── ● decide ok → prepare_run
     ├── ● prepare_run ok → do_run
     └── ◐ do_run tool·balanced
▊  answer here, then Enter…                                                                          ▎
▊  yes                                                                                               ▎
 you: yes
     └── ▶ ◐ do_run tool·balanced
 READY · deepseek · tokens 8244+2273 · session 20260718-013216-8c6d
 ⏸ Consent: machine 'news_search' will call host tool(s) [search] (e.g. live web search). Allow    ▃▃
 agent:
 Here's a brief of this week's news about open-source AI models (July 13–18, 2026):
 └── ▼ ✓ console_agent  done
     ├── ▶ ● do_run ok → decide
     ├── ● decide ok → reply
     └── ● reply ok → END
    administration and AI industry discuss a capability framework keyed to current Chinese
    open‑source model abilities.
    Link
  • I tried China's buzzy free AI. The wait was painful, but the answers impressed me. – Business
    Insider tests China's open‑source GLM‑5.2, finding it matches GPT‑5.5 and Opus 4.8 in context  ▅▅
    window and code tasks.
▊  Now boil that down to a single headline takeaway.                                                 ▎
 RUNNING · deepseek · tokens 8244+2273 · session 20260718-013216-8c6d
    Insider tests China's open‑source GLM‑5.2, finding it matches GPT‑5.5 and Opus 4.8 in context  ▇▇
    Link                                                                                           ▃▃
 ▼ Now boil that down to a single headline takeaway.
 READY · deepseek · tokens 9436+2689 · session 20260718-013216-8c6d
 you: Now boil that down to a single headline takeaway.
 The week’s open‑source AI news is dominated by Chinese models matching or beating US frontier
 systems: Moonshot’s Kimi K3 and GLM‑5.2 both impress on coding and reasoning benchmarks, and the
 Trump administration is discussing a regulatory framework calibrated to those very Chinese
 open‑source capabilities.
▊  /quit                                                                                             ▎