Changelog

• Live RAM and swap stats in the TUI footer with shimmer animation
• Token counter (↑ N) streams live during response generation
• /ram slash command: top processes with proportional bar chart, Metal model estimate for Ollama, and 'other (N)' aggregate
• Working indicator now cycles through status phrases (thinking, weighing options, deciding…)
• RAM pressure now swap-driven: < 1 GB = normal, 1–3 GB = tight, ≥ 3 GB = high
• Process names in /ram normalized from full paths — .app bundle names preferred
• Streaming token callback threaded from Ollama through AgentTurnDeps to the UI

Fixes

• Ollama model weights correctly attributed to Metal GPU buffers (not RSS) in /ram output
• Bar chart in /ram no longer crashes when 'other' total exceeds the largest individual process

• Improved agent planning TUI with step indicators
• Permission prompt redesign with inline confirmation
• Inline working status indicator in the transcript
• /model, /runtime, /context, /permissions slash commands
• edit_file, glob, run_tests, web_fetch tools added to the agent

• Initial release
• llanite install, run, stacks, inspect, doctor, fetch, catalog, search, set, remove, uninstall, clone, models, config
• Ink TUI with transcript, status line, working indicator, and permission prompts
• Model fit detection — reads system RAM and classifies each model candidate
• Local Coder, Repo Reviewer, Private Chat, and Lightweight Coder prebuilt stacks
• Ollama runtime support with automatic model pull on install
• ReAct-style agent with planning modes (auto / on / off) and per-turn tool confirmation