- facts and evidence show their privacy level, the overview shows the level spread in scope,
and level toggles narrow the review list
- fact cards list their sources' extraction tags, the most shared first
- "全选" takes the whole filtered list and confirms when it includes facts not shown yet
- with eight or more domains the clusters alternate between an inner and an outer ring without
overlapping; domain colours follow one OKLCH hue walk
- the synthetic demo profile gains domains, privacy levels and source tags
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- prepare writes peers.json: the chat_with_id and chat_with behind each peer id (P1, P2, ...)
- read_memories returns the peers along with the records
- organized_memories.json carries the peers next to the sources
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- summarize gives every fact a time_scope: stable (never changes), current (latest known state),
superseded (ended, replaced by a later statement, or a one-off claim contradicting well-supported ones)
or dated (a plan, arrangement or event tied to a time)
- a state that ends takes its dependent states and goals with it; finished experiences and place facts are stable
- age, grade and tenure are written as birth year, entry year or start month and labelled stable
- identity (dimension 1) groups of every domain share one summarize request, so the model can see which
candidate a later one replaced
- validation rejects missing or unknown time scopes; the rules sit in the fixed summarize prefix
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- classify gives every memory a sensitivity (public/normal/private/sensitive), including memories with no fields
- summarize and reduce see each source's level and give every fact its own level, lower when the value drops private details
- the level rubric sits in the fixed prompt prefix of classify and summarize; validation rejects missing or unknown levels
- organized_memories sources keep the classify level
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- weclone-cli distill-style extracts speaking-style measurements per person and per contact group
- /api/style serves the newest complete style run to the web 说话风格 view
- distill-style passes --help through to its own option parser
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- organize-memories induces life domains, classifies memories into them, groups attributes into topics and summarizes per topic
- llm-profile-hierarchy splits oversized topics, rewrites domain descriptions and requires --output-dir
- remove canonicalize_state_memories, group_state_memories and organize_identity_attributes
- rename prompts/chat_distill.py to prompts/distill_memory.py; drop merge_enabled from the settings template
- demo profile sources rewritten as casual fictional chats
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- /api/profile/intro picks the atlas night-intro sentences from the profile sources
- review_paths reads only the newest profile_runs/*/profile_hierarchy run
- weclone-cli dev starts the backend and the web dev server together
- make-dataset joins merged messages with a newline
- llm_client sends service_tier=default; auth drops the unused DEFAULT_DATABASE
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- weclone/harness/profile: render the chat, route it to topics (jev
two-stage choice or one LLM call), fill a fact budget and let an LLM
keep the facts the reply needs, in must and helpful tiers. Any failure
leaves the reply without profile facts and records the error.
- Unknown topic or fact ids in an LLM answer are dropped with a warning;
only an answer with no valid id is retried.
- The profile tree applies the review store when its database exists:
rejected facts are dropped and edited attributes and values are used.
- Review store: /api/profile returns the profile's domains (name,
description, topic order).
- Config: harness_args.profile_context.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Integrate suffixed encrypted storage across extraction, review, training, and inference. Add shared-password web authentication with two-hour sessions, private request logging, and an isolated profile-building pipeline.
Fix encrypted dataset discovery at training entrypoints and duplicate CSV imports. Add synthetic storage, authentication, and log checks; update documentation and frontend assets.
- Added default values for batch size, max tokens, and timeout in distill_profile.py.
- Refactored required configuration functions to support section-specific error messages.
- Introduced positive and optional positive integer validation functions for configuration.
- Updated resolve_llm_args to utilize new configuration validation functions.
- Modified organize_memories.py to load agent distill configuration and handle missing model gracefully.
- Improved stage 2 output file management in make_dataset_stage2.py, ensuring proper validation against manifest files.
- Added new web assets including HTML, CSS, and JS files for profile review interface.
- Set up default static directory for serving web assets in profile_review.py.
- Implemented a new authentication system in `auth.py` to manage user sessions and password security.
- Created an `AuthStore` class to handle password hashing, session management, and login/logout functionality.
- Added API endpoints for login, logout, and session validation in the FastAPI application.
- Developed a React context for authentication in `Auth.tsx`, including a login form and logout button.
- Introduced tests for private routes and session management in `test_web_auth.py`.
- Added CSS styles for the authentication page in `index-DOIy3AqG.css`.
- Included a new JavaScript file for the web application in `index-DwlWkAhO.js`.
- Implement comprehensive tests for profile review endpoints, including approval, editing, and exporting facts.
- Validate snapshot import behavior and ensure proper error handling for invalid data.
- Introduce tests for server CLI options and ensure correct application of configurations.
- Test unified routes with lifespan management, authentication, and streaming responses.
- Updated Python version requirement in `pyproject.toml` to 3.12.
- Adjusted dependency specifications for `llamafactory`, `torch`, `torchvision`, and `torchaudio` to include version constraints.
- Updated the `qa_generator.py` to include a new mechanism for managing chat member relationships, allowing the addition of contextual information about the relationship between users in conversations.
- Refactored the CSV loading function to support loading user relationship data from a `users.json` file, improving the context provided during QA generation.
- Added a new configuration option `add_relation` to the dataset settings, enabling users to toggle this feature.
- Updated the `.gitignore` to exclude additional data directories and cache files for better repository hygiene.
- Bumped version to 0.3.03 in `pyproject.toml` to reflect these changes.
Refactors dataset management to consistently append '-vl' for vision-language
datasets and dynamically name cleaned datasets (e.g., 'dataset-cleaned').
Enforces vLLM as a strict dependency for LLM-based data cleaning, exiting if
unavailable. Integrates 'enable_thinking' option for LLM cleaning and enables
cleaning by default in relevant test configurations.
Adds torchvision dependency for vision models and streamlines the cleaning call
in training to centralize decision-making.
Updates the Qwen3 model in test configurations from 4B to the smaller 0.6B version.
This change helps improve test performance and reduce resource requirements.
Ensures the data generation pipeline correctly includes the expected number of <image> tags for multimodal configurations, specifically for Qwen2.5-VL datasets.
Adds a blocked words filter to the dataset creation process.
This prevents undesirable words from appearing in the generated
datasets, enhancing data quality and safety.
A test case is added to verify the filter is working correctly.
Also, skips "text" type in qa_generator to avoid duplicated filter.
Refactors the test environment setup to use pytest fixtures
for session-level setup and teardown. This ensures a clean
and consistent environment before and after each test session.
It also improves the robustness by backing up and restoring
existing 'model_output' and 'dataset/csv' directories.
