2026-03-08 - 2026-09-08
Overview
1 Pull request merged by 1 user
Merged
#61 Refactor application and runtime boundaries
97 Issues closed from 1 user
Closed
#71 add support for a separate vision model (small) to add eyes
Closed
#36 ssd streaming seems to not show enough in the graph
Closed
#99 Optimize and benchmark Qwen3.8 Flash Next on Apple silicon
Closed
#98 Add Qwen3.8 Flash Next native MTP speculation
Closed
#97 Implement exact Qwen Sparse Attention and long-context state
Closed
#96 Stream Qwen3.8 Flash Next PLE from mapped storage
Closed
#95 Implement the Qwen3.8 Flash Next Rust and Metal text core
Closed
#94 Integrate Qwen3.8 Flash Next loading and tokenizer behavior
Closed
#93 Prepare exact Qwen3.8 Flash Next Bare Speed artifacts
Closed
#92 Optional follow-up: add CUDA to the completed Mojo backend
Closed
#91 Complete the macOS Mojo release matrix, packaging, and Metal cutover
Closed
#90 Implement the GLM 5.2 Mojo-Metal backend with isolated sparse-path gates
Closed
#89 Add Qwen3.8 MTP and multimodal inference on Metal without regressing text
Closed
#88 Add a performance-parity Qwen3.8-27B Q4 and Q8 text backend on Metal
Closed
#87 Implement the complete DeepSeek Mojo-Metal backend with performance parity
Closed
#86 Implement the shared Mojo foundation and a quantized Metal vertical slice
Closed
#85 Build the native-Metal-versus-Mojo parity harness and operation inventory
Closed
#84 Spike: prove a license-safe, performance-parity Mojo-Metal integration
Closed
#82 Add a native glob tool for agent file discovery
Closed
#80 Add an inline SVG presentation tool with durable chat replay
Closed
#81 Remove the deprecated DeepSeek V4 Flash preview and legacy MTP
Closed
#83 Make F1 open a complete, up-to-date User Guide
Closed
#55 flow diagrams as widget for A2UI
Closed
#43 "allow always" in the confirmation dialog with project-local storage of allowances
Closed
#79 Add end-to-end MXFP4 GGUF support to the Metal backend
Closed
#76 Add installable agent lifecycle hooks with Ponytail as the reference integration
Closed
#78 Reach current DS4 performance parity on Apple Metal from the Rust execution path
Closed
#75 hiding the sidebar does not enlarge the main area
Closed
#74 Add DeepSeek-style hierarchical AGENTS.md instruction lifecycle
Closed
#70 implement a release workflow that runs on new tags
Closed
#67 support skills in ~/.agents/skills/ and integrate them into the prompt
Closed
#73 Add a Rust-native fresh-agent Ralph loop
Closed
#72 Make agent tool handling schema-driven and recoverable
Closed
#69 ds4 now has support for 0731
Closed
#5 MCP server as additional tools
Closed
#52 integrate OpenSpec into dev-brain
Closed
#68 performance measurements needed to see if we introduced lag
Closed
#66 folders inside the configured dev brain should be treated as "inside the project"
Closed
#35 preferences that are model related need to be stored with the selected model
Closed
#65 automatic approval checks should be visible in history
Closed
#64 thinking mode needs to be selectable
Closed
#63 path is not correctly set in shells inside the app
Closed
#62 add DeepSeek V4 Flash 0731 as a model option
Closed
#54 how is compile of .metal files done in the project?
Closed
#50 rewrite ds4_web.c/.h into rust
Closed
#20 support more search engines
Closed
#51 add a headless web browser for AI web access
Closed
#60 archived sessions should get their kvcache slice delete to make space
Closed
#59 sidebar layout is a bit misaligned looking with threads below projects
Closed
#58 default-diff type (one or two column) should be a setting
Closed
#49 stats should also have a section "current session"
Closed
#45 make the token counts in the chat divider line more readable
Closed
#6 skills as folders inside dev-brain
Closed
#57 heatmap seems to be missing from A2UI
Closed
#56 README does not tell the user where models are stored
Closed
#48 checkboxes in ui should be the macos typical slider widgets and not checkmarks
Closed
#53 references to ../ds4/ files in the project are not allowed due to other people not having ds4 checked out
Closed
#47 we need just one git crate dependency - get rid of gix
Closed
#18 implement SOUL.md and USER.md based on dev brain
Closed
#46 no direct git shell commands in this app, use rust code and gix / gitoxide
Closed
#44 tool output should be by-default cut in output to a limit of 10 lines
Closed
#40 command output should be formatted like a terminal
Closed
#42 the git pane does not show actual diffs when clicking on a file
Closed
#41 every continuation has a "COntext rebuild" block
Closed
#39 add a git pane to the chat / a2ui / stats pill
Closed
#23 show the git branch in the chat area as well as the project we are on
Closed
#33 multiple chats must be able to work in parallel
Closed
#34 scroll handling must be smoother and user-scrolling should not conflict with updates
Closed
#31 session list under projects should be ordered by last use, newest first
Closed
#29 the local agent must read an existing AGENTS.md and add it to the starting prompt of a session
Closed
#8 add a right sidebar that can show different things
Closed
#37 preferences should have a sidebar/main structure with sections
Closed
#30 system prompt needs a multi-line field in preferences and it should be a better default
Closed
#32 deleting a session needs a confirmation dialog
Closed
#38 better rendering of the chat area with blocks and timings and borders
Closed
#28 preferences doesn't allow to tab to next field
Closed
#27 kvcache eviction only addresses transient cache and limits should only hit that, but the bar chart needs to list sessions as a bucket
Closed
#26 move preferences from database to yaml file
Closed
#19 context fill grade should be a little pie chart instead of a x/y text
Closed
#3 cache management for kvcache disc usage
Closed
#25 the divider between sidebar and chat/stats area should be draggable
Closed
#21 settings should have descriptive hovertext when mouse is over the title of the setting
Closed
#22 clicking the name of a project should collapse/expand it
Closed
#24 opening the app should select the last project we were on automatically and open a draft chat for that project
Closed
#7 add icons to collaps/expand the sidebar
Closed
#17 Preference save silently discards the runtime update on a poisoned lock; poisoning conventions are inconsistent
Closed
#16 Tokenizer panics on non-char-boundary offsets where the ds4 original degrades gracefully
Closed
#15 build.rs does not track native/metal headers; stale object file can be linked after a header change
Closed
#14 Executor field order is load-bearing for teardown safety and undocumented
Closed
#13 Buffer::view has no lifetime tie to its base buffer
Closed
#12 A panic on the model runtime thread permanently disables generation
Closed
#11 HTTP endpoint spawns unbounded threads and has no total request or write deadline
Closed
#10 Model download has no network timeouts; a stalled server hangs the download thread forever
Closed
#9 Client disconnect does not cancel generation; GPU keeps running for abandoned requests
Closed
#4 session title should be set by the AI on first chat
Closed
#1 "..." on the session title should provide a menu of quick actions on sessions
Closed
#2 statistics for prefill seem not to properly update
99 Issues created by 1 user
Opened
#1 "..." on the session title should provide a menu of quick actions on sessions
Opened
#2 statistics for prefill seem not to properly update
Opened
#3 cache management for kvcache disc usage
Opened
#4 session title should be set by the AI on first chat
Opened
#5 MCP server as additional tools
Opened
#6 skills as folders inside dev-brain
Opened
#7 add icons to collaps/expand the sidebar
Opened
#8 add a right sidebar that can show different things
Opened
#9 Client disconnect does not cancel generation; GPU keeps running for abandoned requests
Opened
#10 Model download has no network timeouts; a stalled server hangs the download thread forever
Opened
#11 HTTP endpoint spawns unbounded threads and has no total request or write deadline
Opened
#12 A panic on the model runtime thread permanently disables generation
Opened
#13 Buffer::view has no lifetime tie to its base buffer
Opened
#14 Executor field order is load-bearing for teardown safety and undocumented
Opened
#15 build.rs does not track native/metal headers; stale object file can be linked after a header change
Opened
#16 Tokenizer panics on non-char-boundary offsets where the ds4 original degrades gracefully
Opened
#17 Preference save silently discards the runtime update on a poisoned lock; poisoning conventions are inconsistent
Opened
#18 implement SOUL.md and USER.md based on dev brain
Opened
#19 context fill grade should be a little pie chart instead of a x/y text
Opened
#20 support more search engines
Opened
#21 settings should have descriptive hovertext when mouse is over the title of the setting
Opened
#22 clicking the name of a project should collapse/expand it
Opened
#23 show the git branch in the chat area as well as the project we are on
Opened
#24 opening the app should select the last project we were on automatically and open a draft chat for that project
Opened
#25 the divider between sidebar and chat/stats area should be draggable
Opened
#26 move preferences from database to yaml file
Opened
#27 kvcache eviction only addresses transient cache and limits should only hit that, but the bar chart needs to list sessions as a bucket
Opened
#28 preferences doesn't allow to tab to next field
Opened
#29 the local agent must read an existing AGENTS.md and add it to the starting prompt of a session
Opened
#30 system prompt needs a multi-line field in preferences and it should be a better default
Opened
#31 session list under projects should be ordered by last use, newest first
Opened
#32 deleting a session needs a confirmation dialog
Opened
#33 multiple chats must be able to work in parallel
Opened
#34 scroll handling must be smoother and user-scrolling should not conflict with updates
Opened
#35 preferences that are model related need to be stored with the selected model
Opened
#36 ssd streaming seems to not show enough in the graph
Opened
#37 preferences should have a sidebar/main structure with sections
Opened
#38 better rendering of the chat area with blocks and timings and borders
Opened
#39 add a git pane to the chat / a2ui / stats pill
Opened
#40 command output should be formatted like a terminal
Opened
#41 every continuation has a "COntext rebuild" block
Opened
#42 the git pane does not show actual diffs when clicking on a file
Opened
#43 "allow always" in the confirmation dialog with project-local storage of allowances
Opened
#44 tool output should be by-default cut in output to a limit of 10 lines
Opened
#45 make the token counts in the chat divider line more readable
Opened
#46 no direct git shell commands in this app, use rust code and gix / gitoxide
Opened
#47 we need just one git crate dependency - get rid of gix
Opened
#48 checkboxes in ui should be the macos typical slider widgets and not checkmarks
Opened
#49 stats should also have a section "current session"
Opened
#50 rewrite ds4_web.c/.h into rust
Opened
#51 add a headless web browser for AI web access
Opened
#52 integrate OpenSpec into dev-brain
Opened
#53 references to ../ds4/ files in the project are not allowed due to other people not having ds4 checked out
Opened
#54 how is compile of .metal files done in the project?
Opened
#55 flow diagrams as widget for A2UI
Opened
#56 README does not tell the user where models are stored
Opened
#57 heatmap seems to be missing from A2UI
Opened
#58 default-diff type (one or two column) should be a setting
Opened
#59 sidebar layout is a bit misaligned looking with threads below projects
Opened
#60 archived sessions should get their kvcache slice delete to make space
Opened
#62 add DeepSeek V4 Flash 0731 as a model option
Opened
#63 path is not correctly set in shells inside the app
Opened
#64 thinking mode needs to be selectable
Opened
#65 automatic approval checks should be visible in history
Opened
#66 folders inside the configured dev brain should be treated as "inside the project"
Opened
#67 support skills in ~/.agents/skills/ and integrate them into the prompt
Opened
#68 performance measurements needed to see if we introduced lag
Opened
#69 ds4 now has support for 0731
Opened
#70 implement a release workflow that runs on new tags
Opened
#71 add support for a separate vision model (small) to add eyes
Opened
#72 Make agent tool handling schema-driven and recoverable
Opened
#73 Add a Rust-native fresh-agent Ralph loop
Opened
#74 Add DeepSeek-style hierarchical AGENTS.md instruction lifecycle
Opened
#75 hiding the sidebar does not enlarge the main area
Opened
#76 Add installable agent lifecycle hooks with Ponytail as the reference integration
Opened
#77 Explore zero-copy mmap-backed expert paging
Opened
#78 Reach current DS4 performance parity on Apple Metal from the Rust execution path
Opened
#79 Add end-to-end MXFP4 GGUF support to the Metal backend
Opened
#80 Add an inline SVG presentation tool with durable chat replay
Opened
#81 Remove the deprecated DeepSeek V4 Flash preview and legacy MTP
Opened
#82 Add a native glob tool for agent file discovery
Opened
#83 Make F1 open a complete, up-to-date User Guide
Opened
#84 Spike: prove a license-safe, performance-parity Mojo-Metal integration
Opened
#85 Build the native-Metal-versus-Mojo parity harness and operation inventory
Opened
#86 Implement the shared Mojo foundation and a quantized Metal vertical slice
Opened
#87 Implement the complete DeepSeek Mojo-Metal backend with performance parity
Opened
#88 Add a performance-parity Qwen3.8-27B Q4 and Q8 text backend on Metal
Opened
#89 Add Qwen3.8 MTP and multimodal inference on Metal without regressing text
Opened
#90 Implement the GLM 5.2 Mojo-Metal backend with isolated sparse-path gates
Opened
#91 Complete the macOS Mojo release matrix, packaging, and Metal cutover
Opened
#92 Optional follow-up: add CUDA to the completed Mojo backend
Opened
#93 Prepare exact Qwen3.8 Flash Next Bare Speed artifacts
Opened
#94 Integrate Qwen3.8 Flash Next loading and tokenizer behavior
Opened
#95 Implement the Qwen3.8 Flash Next Rust and Metal text core
Opened
#96 Stream Qwen3.8 Flash Next PLE from mapped storage
Opened
#97 Implement exact Qwen Sparse Attention and long-context state
Opened
#98 Add Qwen3.8 Flash Next native MTP speculation
Opened
#99 Optimize and benchmark Qwen3.8 Flash Next on Apple silicon
Opened
#100 support deepseek v4 flash vision experimental