commits
- This is due to the cursor is not moved after that and
so it could be override. So we explicitly add double newline to
move cursor down.
- It was due to improper handling of assistant msg
- models_used field in SharedSession is a list
- Each chats will have the model name it used, since ideally a session
can have multiple models used
* Update CLI ASCII art and product copy
* Align CLI product description copy
* Format CLI copy assertion
* wip: tried atrium for atproto login
* wip: generated auth url for atproto w bluesky
- nxt callback and saving the tokens, then write records
* wip: Done with callback processing and session retrival
- nxtup: Persist the session by impl SessionStore
* feat: At proto login/logout complete with commands
Update the installer welcome text and README tagline to use the same Tiles positioning language.
Made-with: Cursor
Use the same "Install Tiles" title for both network and offline pkg installers to keep installer branding consistent.
Made-with: Cursor
Use theme-specific Tiles logo backgrounds in both network and offline installer distributions and refresh the first intro paragraph to local-first messaging for consistency with current product positioning.
Made-with: Cursor
- pi is bun binary so has JIT so need to explicitly tell apple
to allow JIT via entitlments, else we can dynamic memory allocation error
while executing Pi
fix: sending SSE events with typed events as openAI
fix: Update CLI model response rendering wrto new SSE response changes
fix: weabing more standards for model response so that its rendered on PI
- TODO: ofcourse Rust CLI repl rendering is broke, need to fix it next
wip: packaging pi
wip: packaging pi
feat: Calling pi from tiles as repl
- exposes a GET model_cache_path API for py server to load correct model
feat: added current model in config.toml + build script for pi
- having current model in config.toml allows pi to read it from there
and update its models.json
- build script for pi for DEV, so dev can run updated pi binary
feat: Added -p flag for PI repl + Daemon hot reloading
- PI repl will be under `-p` flag for now. `tiles -p` or `tiles run modelfile -p`
- FIX: On using different versions of Tiles, the bg daemon will also reload to the
correct version
feat: Able to hook into pi rpc , added /state command
- nextup steaming agent responses
Updated the status description for clarity and emphasis on exploratory use.
Added status information indicating that Tiles is in alpha quality and may have bugs.
Updated descriptions to emphasize local-first privacy and clarified the mission and purpose of Tiles.
health command upgrade and other QOL changes
- fix: proper recursive exit after mdel download to load the model
- fix: proper recursive exit after mdel download to load the model
fix: config + user folder creation preceeds ftue
- Since the streamlining of db connection, it needs to have
config, user dirs available. But we only creates then in FTUE.
So now we made these folder creations as part of core init
streamline db conn references
fix: updated mlx related pkgs + end condition for stopping model request loop
- pre-install script to remove prev server folder by the installer
- removed iroh error logs showing in release build
- removed HF deps as downloader is done via RUST
- Will abort the sync if there is any write err, so that resync or debug
is possible
Added encryption at rest for the DBs
- row_counter field will be used as a kind of version vector for p2p chat sync
P2P device linking v1
- We now verify the DID coming from the msg can be derived from the sender's Public key
- Instead of hardcoded topic for every session, from now it will be unique topics (online mode only)
- refactor
- Build changes for 0.4.5
Local device linking with mDNS
- Added mDNS peer discovery with hex shortcode of 8 digit
- Added link utility commands such as link disable <DID> and link list-peers
- Refactored such that linking with tickets and codes coexist
fix: sending SSE events with typed events as openAI
fix: Update CLI model response rendering wrto new SSE response changes
fix: weabing more standards for model response so that its rendered on PI
- TODO: ofcourse Rust CLI repl rendering is broke, need to fix it next
wip: packaging pi
wip: packaging pi
feat: Calling pi from tiles as repl
- exposes a GET model_cache_path API for py server to load correct model
feat: added current model in config.toml + build script for pi
- having current model in config.toml allows pi to read it from there
and update its models.json
- build script for pi for DEV, so dev can run updated pi binary
feat: Added -p flag for PI repl + Daemon hot reloading
- PI repl will be under `-p` flag for now. `tiles -p` or `tiles run modelfile -p`
- FIX: On using different versions of Tiles, the bg daemon will also reload to the
correct version
feat: Able to hook into pi rpc , added /state command
- nextup steaming agent responses