Release v0.5.0

Changelog

[Unreleased]

[0.5.0] - 2026-05-18

Added

• Async Claude Code (Codex) agent sessions: Elroy can now launch and manage Claude Code coding sessions as background agents, with results surfaced back in the TUI. Use the new tools to create and track sessions from within a conversation
• File reading tool: Elroy can now read files from the local filesystem as part of tool use, enabling richer context for coding and research tasks
• Feature request flow: Users can now submit feature requests directly from the TUI; these are tracked and surfaced to Elroy for prioritization
• Self-reflection improvement items are now surfaced in the TUI sidebar, giving visibility into what Elroy is tracking about its own behavior

Improved

• The TUI sidebar now supports keyboard navigation between tabs using arrow keys
• Ctrl-C now clears the chat input buffer rather than exiting, making it easier to cancel a partially typed message
• Multi-line paste support for the chat input
• Long messages now wrap correctly in the input area
• Context from previous sessions loads more reliably when starting a new chat
• Internal thought rendering has been fixed to display correctly in the conversation view

Changed

• Slash commands have been removed from the TUI. Actions previously available as slash commands are now handled through Elroy's tool system or the sidebar UI directly
• The user ID layer has been removed from the internal context model, simplifying session and repository logic throughout the codebase

Infrastructure

• Documentation is now self-hosted

[0.4.0] - 2026-04-03

Added

• A full-screen Textual terminal UI is now the primary Elroy experience, with live streaming responses, slash-command completion, and a status bar that shows what Elroy is doing in real time
• A browsable sidebar for memories and agenda items, including keyboard navigation, detail views, and inline actions like completing or deleting due items
• File-backed agenda items with checklist tracking, update history, and optional time- or context-based triggers
• Background document ingestion and file sync for local knowledge and memory files

Improved

• Memories and agenda items are now much more transparent in the UI: you can inspect what Elroy has surfaced, open individual items, and act on them directly
• Due items now appear inline in the agenda view instead of living in a separate panel, reducing duplication between reminders and task tracking
• Status feedback during chat is more granular, with clearer visibility into loading context, recalling memories, running tools, and other long-running steps
• Memory storage is now file-backed, which makes local data easier to inspect, manage, and sync with other tools
• Startup and chat behavior are more deterministic and responsive, with better autocomplete, more consistent browsing controls, and improved local test coverage for the TUI
• Dependencies are now pinned more tightly, which should make installs and upgrades more reproducible across environments

Changed

• Reminders have been folded into the agenda model as due items, so timed and contextual follow-ups now live alongside regular agenda work instead of in a separate system
• Elroy is now focused on a local terminal workflow rather than supporting multiple product surfaces
• Configuration is centered on config files and environment variables, with the TUI as the main entry point

Removed

• The web API / HTTP interface has been removed
• Postgres and ChromaDB backends have been removed; Elroy now uses the local SQLite-based stack only
• Older CLI-only flows and related update/install paths that no longer fit the TUI-first product direction have been removed

[0.3.0] - 2026-01-25

Added

• ChromaDB integration: Elroy now supports ChromaDB as a vector database backend for memory storage, providing an alternative to SQLite with improved vector search capabilities. Configure with --chroma-db or set chroma_db: true in your config
• Migration script for transitioning existing SQLite memories to ChromaDB (scripts/migrate_to_chroma.py)
• Claude Code skills integration: Install Elroy memory tools as Claude Code slash commands (/remember, /recall, /list-memories, /remind, /list-reminders, /ingest)
• New elroy install-skills CLI command for easy installation of Claude Code skills
• New just install-claude-skills and just uninstall-claude-skills commands for convenient skills management
• Comprehensive documentation for Claude Code integration in claude-skills/README.md
• Comprehensive latency tracking system with per-request breakdown of operations, helping identify performance bottlenecks
• Request IDs for tracking operations across the full message processing pipeline
• Detailed logging for all major operations including context loading, memory recall, LLM calls, tool execution, and context persistence

Improved

• Significantly improved performance visibility with operations >100ms logged at INFO level
• Added timing instrumentation to memory consolidation (now logs query count and duration)
• Better structured logging with operation-specific metadata and latency summaries
• Embedding API calls now log character count and retry attempt information

Infrastructure

• Enhanced release automation script with comprehensive error handling and recovery mechanisms
  • Fixed IndexError when checking remote tags by properly handling empty output
  • Changed from lightweight to annotated tags following Git best practices (includes tagger, date, and message)
  • Added backward compatibility for checking both annotated and lightweight tag formats
  • Wrapped release workflow in error handling with automatic cleanup on failure
  • Added capability to delete failed release branches for cleaner repository state
  • Made tag operations idempotent to safely retry after partial failures
  • Fixed write_release_notes.sh to use current version instead of calculating next version

[0.2.0] - 2026-01-22

Added

• Strong and weak model configuration support, allowing you to configure different models for different types of operations for better cost/performance tradeoffs

Improved

• Context message handling now avoids unnecessary rewrites, improving performance and reducing API calls
• Consolidated session context gathering for more efficient memory operations

Infrastructure

• Database migrations for improved index performance on vector queries
• Refactored context message operations and validation logic

[0.1.2] - 2025-12-14

Fixed

• Fixed startup issue with Click dependency that prevented the CLI from launching properly

Improved

• Limited the number of memories displayed in the context panel for better readability and reduced clutter

[0.1.1] - 2025-12-03

Improved

• Updated default model versions for better performance and reliability
• Document ingestion now ignores dotfiles by default to prevent ingesting system files
• Documents are no longer re-ingested if they are moved within the same directory structure

Fixed

• Fixed display bug in memory consolidation interface

Infrastructure

• Refactored LLM client to class-based architecture for better maintainability
• Streamlined configuration management
• Removed MCP server functionality (moved to separate package)
• Removed shell command tools from developer tools

[0.1.0] - 2025-08-16

Added

• Major reminders system replacing goals for better task and event management
• Waitlist functionality for user management
• New document ingestion improvements with better memory consolidation

Improved

• Enhanced memory parsing and consolidation system
• Updated context window to 100k tokens by default for better conversation handling
• Better error handling for API endpoints
• Improved CLI interface and user experience

Infrastructure

• Replaced system messages with synthetic tool calls for better LLM interaction
• Database schema updates for unified reminders system
• Enhanced testing and error recovery mechanisms

[0.0.82] - 2025-08-05

Added

• Roadmap doc
• Web API with endpoints for memory augmentation and querying
• User ID support for vector storage, enabling multi-user memory isolation
• Root API route for health checks and service status

Improved

• Enhanced CLI header display with better formatting
• Updater now supports uv tool installation with fallback to pip for more reliable package management
• Memory augmentation operations now available through API endpoints

Infrastructure

• Added web API infrastructure with FastAPI integration
• Database migrations for user ID support in vector storage tables

[0.0.81] - 2025-07-10

Added

• Status table for document ingestion in directories, with improved real-time feedback

Improved

• Major performance improvements for simple exchanges by adding a decision step for whether memory check is needed
• Irrelevant memories are now ejected asynchronously, reducing context noise and improving recall
• Document ingestion logic updated: newly ingested docs are no longer added to status, and status is shown for directory ingests
• Turned off assistant greeting by default for a less intrusive startup experience
• Do not consider messages starting with /ask as slash commands
• Memory filtering on startup is now removed for faster initialization

Fixed

• Bugfixes for memory ejection and context refresh logic
• More robust handling of document ingestion and status reporting
• Various minor bugfixes in async task scheduling and memory operations

Infrastructure

• Delayed import of scipy and DBSCAN to speed up startup and avoid unnecessary dependencies
• Cleaned up and standardized error messages
• Updated and expanded test coverage for memory and document ingestion features

[0.0.80] - 2025-06-14

Improved

• Enhanced memory filtering with relevance processing to reduce noise
• Updated default log level

Breaking change

• Removed Aider integration. The integration caused dependency issues

[0.0.79] - 2025-04-20

Added

• Add opt-in shell execution tool

Improved

• More configurable tool and display options:
  • Option to exclude memory panel display
  • More granular dependencies

[0.0.78] - 2025-04-10

Added

• Simplified memory retrieval assistant tool: Assistant tool can now better retrieve source document information for memories
• Ensure a memory is created at least every n messages (configurable)
• Plaintext output option for messages

Fixed

• Tool call argument validation
• API key requirement handling
• Message timestamp consistency
• Context message validation

Improved

• Default user name changed from "UNKNOWN" to "User"
• PostgreSQL configuration flexibility

[0.0.77] - 2025-03-16

Added

• Support for Arize-Phoenix tracing!
• Directory-based document ingestion with recursive and pattern matching: try it with elroy ingest
• Reflective memory option: Increase reflection of memories
• Overhauled documentation

Fixed

• Fix for context overflow in embeddings
• Context message ordering bug
• Assistant name consistency in memory operations

Improved

• Shorter, more focused greeting messages
• More engaging persona and tone

[0.0.76] - 2025-02-28

Added

• Added ingest_doc API function
• Memory update functionality with timestamp tracking
• Example notebook for repository document ingestion

Fixed

• Context message concurrent operations handling
• Missing timestamp in context messages
• Bug where /pop function ejected too many messages operation

Improved

• Installation documentation updates for stable branch

[0.0.75] - 2025-02-23

Added

• Document ingestion support for markdown files!
• New tools for examining memory sources and origins

Improved

• Add --first flag to disable assistant greeting
• Enhanced memory retrieval with source tracking

[0.0.74] - 2025-02-18

Added

• Support for configuring OpenAI API base URL and API key
• Remember function restored to API

Fixed

• More reliable handling of empty messages
• Better string conversion for text outputs

Improved

• Enhanced inline tool handling with validation

[0.0.73] - 2025-02-15

Added

• Automatic model selection based on available API keys:
  • With ANTHROPIC_API_KEY: Uses Claude 3 Sonnet
  • With OPENAI_API_KEY: Uses GPT-4o and text-embedding-3-small
  • With GEMINI_API_KEY: Uses Gemini 2.0 Flash
• Display of selected models at startup for better transparency

Fixed

• Improved handling of new user creation in API mode

[0.0.72] - 2025-02-13

Added

• /save command

Improved

• Updated logic for selecting chat and embedding model API
• Support for Gemini models

[0.0.71] - 2025-02-07

Added

• New slash commands!
  • Accessing memories and goals:
    • /search_memories: Search against goals and memories
    • /print_active_goals
    • /print_complete_goals
    • /print_memories
  • Managing context messages:
    • /pop: Remove n messages from convo
    • /rewrite: Rewrite the last assistant response
• New API function: record_message, useful for backfilling (see examples/backfill_chat_logs.py)
• Discord bot script

Fixed

• Clearer error handling for SQLite platform mismatch
• Slash commands can now be cancelled

[0.0.70] - 2025-02-03

Added

• Code syntax highlighting for code in Elroy responses
• New remember_convo command for conversation memory
• New CLI command: list-tools
• Installation script for easier setup

Infrastructure

• Small updates on top of 0.0.69

[0.0.69] - 2025-02-03

Added

• Code syntax highlighting for code in Elroy responses
• New remember_convo command for conversation memory
• New CLI command: list-tools
• Installation script for easier setup

Improved

• More graceful handling of invalid parameters with warnings
• Documentation updates and clarifications

[0.0.68] - 2025-01-27

Added

• MCP support! Use elroy mcp print-config for the server configuration

[0.0.67] - 2025-01-26

Added

• API expanded to include more memory and goal functions

[0.0.66] - 2025-01-26

Infrastructure

• Migrated from Poetry to UV for streamlined package management

[0.0.65] - 2025-01-24

Fixed

• Resolved threading issues in sqlite for improved stability
• Fixed tool calling schema injection for better reliability

Improved

• Assistant name now displays correctly in title bar

[0.0.64] - 2025-01-23

Fixed

• Improved error recovery and differentiation in toolkit
• --inline-tools option for models that do not natively support tool calls.

Improved

• Context refresh is now based on context token counts, rather than wall clock time

[0.0.63] - 2025-01-22

Fixed

• Improved release script reliability and deployment consistency

[0.0.62] - 2025-01-21

Added

• Support for custom user-defined tools
• Stream internal thought output
• Option to show sensitive values in print-config command

Fixed

• Improved timezone handling in login messages
• More robust tool calling
• Fixed print-config functionality

[0.0.61] - 2025-01-10

Fixed

• Improved handling of system paths

[0.0.60] - 2025-01-09

Improved

• Added some debug info to print-config

[0.0.59] - 2025-01-06

Improved

• print-config now works without requiring database connection for basic operations like viewing settings

[0.0.58] - 2025-01-05

Improved

• Improve readability of /print_config and /print_context_messages commands
• Bugfixes to declutter conversation history

[0.0.57] - 2025-01-04

Added

• Memory consolidation overhaul: Consolidation is now based on how many memories have been created since the last consolidation operation, rather than time.
• Consolidated memoires now store source metadata
• Maximum limit on consecutive tool calls to prevent infinite loops

Improved

• Commands are now regular CLI commands, rather than flags
• Autocomplete improvements
• Results from assistant function calls are now printed to the console

[0.0.56] - 2024-12-30

Improved

• Fixes for command autocomplete

[0.0.55] - 2024-12-29

Added

• Custom assistant name configuration via new command
• Default configuration file location for easier setup
• UV package manager installation instructions

Improved

• Enhanced conversation context management and summarization
• Streamlined database setup documentation
• More comprehensive conversation summaries including tool interactions
• Enable or disable tools via explicit parameter rather than introspection of model settings (enabled by default)

Documentation

• Expanded README with additional installation options
• Clearer database setup instructions

[0.0.54] - 2024-12-26

Added

• SQLite support! The defualt data store is now SQLite.

Breaking changes

• The default data store is now SQLite, and the database env var has been renamed from ELROY_POSTGRES_URL to ELROY_DATABASE_URL (ELROY_POSTGRES_URL will still be recognized for backward compatibility)

[0.0.53] - 2024-12-19

Improved

• New /help command with improved command autocomplete functionality
• Cross-platform support for Elroy cache and home directories using platformdirs
• ctrl-r Conversation search now spans sessions

Infrastructure

• DB schema migrations to prepare for sqlite support
• Added Discord community server

Documentation

• Added Discord server link to README
• Update Docker setup documentation

[0.0.52] - 2024-12-16

Added

• Ability to create multiple "users" in a single deployment. Each user has isolated memories.
• Customizable personas: Specify your own default persona for all assistants via the default_persona config value, or use --set-persona to set a persona for one specific assistant.
• Model alias support for easier model selection: --o1, --gpt-4o, --sonnet, etc. See README for full list of aliases.
• Configurable assistant greeting: use no-enable-assistant-greeting to prevent the assistant from sending the first message.

Improved

• Enhanced model fallback behavior: Primarily motivated by Anthropic - if a rate limit is used, Elroy will automatically attempt to switch to a fallback model (order of fallback behavior can be found via elroy --list-models)
• Better detection of tool support for OpenAI and Anthropic models: Chat functionality and memory consolidation are now supported for models that do not support tools.
• Streamlined CLI commands and tools organization

Fixed

• Various minor bugfixes and stability improvements

[0.0.51] - 2024-11-30

Improved

• Fix multi-argument system commands. For user initiatied system commands with multiple arguments, the user will now be prompted to enter each argument.

[0.0.50] - 2024-11-28

Breaking changes

• CLI commands have been converted to flags, so:
  • remember is now --remember
  • remember with -f is not --remember-file
  • list-models is now --list-models
  • chat is now --chat. This remains the default command if no command is provided.
• Docker integration is updated
  • --use-docker-postgres flag is removed. You can start the docker database using the scripts/docker_postgres.py script.

Improved

• Enhanced CLI configuration with streamlined flag organization
• Optimized default parameters for better out-of-box experience
• Improved Docker integration for more reliable container usage

Fixed

• Resolved various test suite issues

[0.0.49] - 2024-11-25

Improved

• Replaced JSON LLM outputs with Markdown, added more fault tolerance in parsing responses with unexpected formatting.

[0.0.48] - 2024-11-22

Fixed

• Fixed type mismatch in context refresh wait timing

Infrastructure

• Updated Discord release announcements to use announcements channel

[0.0.47] - 2024-11-22

Added

• Added configurable LLM response caching: enable_caching, defaulting to true.
• New system commands for troubleshooting:
  • /tail_elroy_logs: View log Elroy logs from within the chat UI
  • /print_elroy_config: View Elroy config from within the chat UI
  • /create_bug_report: Open a pre-filled bug report in the browser (available to the user only, not the assistant)

[0.0.46] - 2024-11-20

Added

• Added expanded configuration options for better customization

Fixed

• Improved CHANGELOG.md handling with better path resolution and error management

[0.0.45] - 2024-11-19

Added

• Added automated Discord announcements for new releases

Infrastructure

• Improved CI workflow to prevent duplicate runs on branch pushes

[0.0.44] - 2024-11-19

Improved

• Enhanced release process with streaming subprocess output and progress logging
• Updated documentation for clarity and completeness

Legal

• Changed project license to Apache 2.0

[0.0.43] - 2024-11-18

Fixed

• Minor fixes

[0.0.42] - 2024-11-17

Added

• Updated README to document all startup options and system commands for better user guidance.
• Added more verbose error output for tool calls to improve debugging and error tracking.

Fixed

• Improved autocomplete functionality by filtering goals and memories for more relevant options.
• Simplified demo recording script for easier demonstration creation.

Improved

• Enhanced error handling for goal-related functions to better surface available goals.
• Added override parameters to name setting functions to discourage redundant calls.
• Provided additional context in login messages for a more informative user experience.

Infrastructure

• Added a wait-for-pypi job to verify package availability before Docker publishing, ensuring smoother deployment processes.

[0.0.41] - 2024-11-14

Infrastructure

• Updates to package publishing

[0.0.40] - 2024-11-14

Added

• Initial release of Elroy, a CLI AI personal assistant with long-term memory and goal tracking capabilities.
• Features include long-term memory, goal tracking, and a memory panel for relevant memories during conversations.
• Supports installation via Docker, pip, or from source.
• Includes commands for system management, goal management, memory management, user preferences, and conversation handling.