v0.91.5

What's Changed

MR: #340
Full Changelog: v0.91.4...v0.91.5

Addresses:
#306
#335
#339

Includes:

• Unified ContextManager tool for adding and removing files from context
• #341 - Standardize where /save and /load save files
• #328 - Add --show-speed to see response generation time metrics

New Contributors

• @gopar made their first contribution in #341
• @sannysanoff made their first contribution in #328

v0.91.4

What's Changed

MR: #338
Full Changelog: v0.91.3...v0.91.4

Addresses:
#326
#335
#336
#337

v0.91.3

What's Changed

MR: #331
Full Changelog: v0.91.2...v0.91.3

Addresses:
#330

Includes:

• #319 - Docker container set up for nvidia hardware for local LLMs
• #324 - Add openai compatible provider automatic lookup (currently, streaming must be disabled to use the non-default providers)
• #327 - Fix test suite for main.py, start converting test suite to be pytest as opposed to unittest-based
• Carry over MCP servers, load Local tools for agent mode on switching coder type

New Contributors

• @minger0 made their first contribution in #319

v0.91.2

What's Changed

MR: #321
Full Changelog: v0.91.1...v0.91.2

Addresses:
#320

Includes:

• Refactor prompt system to load from yml files (so eventually custom prompts can just be written as yml files)
• Proper input text selection highlighting
• /help {message} command actually sending the docs to the LLM for user assistance

v0.91.1

What's Changed

MR: #315
Full Changelog: v0.91.0...v0.91.1

Includes:
#303
#314
#318

Addresses:

• Fix tool call output formatting with the little "ᴸ"
• Identify assistant responses in TUI more clearly
• More repo map processing optimizations (more caching, slightly more memory efficient)
• File stubs when /context-management enabled to give models a view of the shape of a file without committing to the whole contents
• Delay completion menu highlighting until a selection is made

v0.91.0

What's Changed

MR: #311
Full Changelog: v0.90.6...v0.91.0

Addresses:
#310

Includes:

• (non-user facing) commands.py decomposition into a base class and registrar of discrete command files
• #305 - Preserve trailing newlines in InsertBlock tool call
• #307 - TUI enhancements, open editor with ctrl+o from TUI
• #312 - Update benchmark harness

v0.90.6

What's Changed

MR: #299
Full Changelog: v0.90.5...v0.90.6

Addresses:
#296
#297
#298
#300
#301

Includes:

• #294 - Copypaste documentation for non-api based usage
• #302 - Handle pre-start up confirmations for TUI
• Get file watcher working in TUI mode for --watch-files
• Get voice input with /voice working again generally (at least on windows, y'all can figure out linux audio drivers for sounddevice)

New Contributors

• @DhirajBhakta made their first contribution in #302

v0.90.5

What's Changed

MR: #291
Full Changelog: v0.90.4...v0.90.5

Addresses:
#290
#292

Includes:

• Prefix completion actually populates in input area (no more needing an extra key press as you tab through)
• Giving LLMs more leeway to react to their own errors
• Specifying --context-compaction-max-tokens as a number less than 1 make it a ratio of the models max context size

v0.90.4

What's Changed

MR: #282
Full Changelog: v0.90.3...v0.90.4

Addresses:
#166
#280
#281
#286

Includes:

• Replace networkx with rustworkx for repo map calculation time optimization
• Make TUI default for non linear-output mode
• #276 - Add copy-paste mode with cp: prefix for web UI interaction
• #283 - Documentation updates
• #284 - Auto-commit in agent mode on finished tool call
• #285 - Fix bugs when running the Git tools

New Contributors

• @chrisnestrud made their first contribution in #283

v0.90.3

What's Changed

MR: #279
Full Changelog: v0.90.2...v0.90.3

Includes:

• Support for gemini-3-flash and github copilot models
• #274 - Todo list preserved in session manager
• #275 - Fix tool calls with mutually exclusive parameters
• #278 - Fix using models with colons in model name