Skip to content

JIN Core Engine v1.0 - #29

Merged
makeitdouble merged 168 commits into
masterfrom
feature/L4_memory_layer
Oct 1, 2026
Merged

makeitdouble merged 168 commits into
masterfrom
feature/L4_memory_layer

Conversation

@makeitdouble

Copy link
Copy Markdown
Owner
  • Introduced the Long-Term Memory pipeline with idle extraction, consolidation, merge/update flows, provenance tracking, tombstones, deduplication, and cross-session persistence.
  • Added L-T memory conflict resolution, fact replacement, absorbed/anchor fact handling, recall actions, and mention-based recall decay.
  • Added dynamic L-T lazy loading based on context pressure and available context-window budget.
  • Made Active Memory autonomous and simplified its runtime contract, update flow, sorting, and persistence.
  • Added persistent Delayed Memory reports with append, pin, restore, edit, tagging, load/unload, and file-backed storage.
  • Added memory tracing, fact lifecycle metadata, merge diffs, reference previews, hover inspection, and editable memory metadata.
  • Expanded reasoning citations across FRAME, Active, Delayed, L-T, files, and runtime state with synchronized UI and Live Avatar highlighting.
  • Reworked the memory panel into dedicated memory tabs with improved deletion, restore, pinning, scrolling, previews, and long-tap interactions.
  • Added the structured runtime context viewer with collapsible blocks, action history, memory state, runtime settings, and context usage.
  • Unified runtime actions around shared contracts, parsing, validation, guards, lifecycle state, failures, follow-ups, and UI representation.
  • Refactored the runtime-action dispatcher into dedicated modules and removed duplicated, obsolete, and legacy action-handling paths.
  • Added malformed-action recovery, bare-action fallback handling, duplicate execution protection, marker validation, and stronger action parsing.
  • Improved confirmations, cancellation, interruption, reconnect replay, stale-action recovery, and pending-action handling.
  • Hardened response and reasoning stream validation against repeated sentences, ASCII drift, malformed markers, empty actions, and infinite loops.
  • Preserved reasoning across turns and follow-ups, added reasoning formatting, citation previews, and improved reasoning-history restoration.
  • Added persistent file attachments with stable IDs, attach/load flows, previews, context injection, delayed-memory integration, and file actions.
  • Added folder reading, file lookup by ID, project/file skills, runtime asset handling, and tool-result cleanup by ID.
  • Added Deep Web Search orchestration with sub-agents and runtime integration.
  • Added the generic MCP client, CALL_MCP, Streamable HTTP/SSE/stdio support, and the Blender MCP skill.
  • Added the agent Posting Board skill and expanded the skill runtime with persistent skill state and reusable tool flows.
  • Expanded the Live Avatar with memory rings, context-pressure indication, reasoning animation, runtime-change markers, reactions, size/color/position/speed controls, and memory-driven effects.
  • Refined memory-attention/metabolism behavior and separated memory-attention logic from obsolete chemical/metabolism experiments.
  • Added JIN bubble skins, theme refinements, light-theme fixes, layout cleanup, panel behavior improvements, and extensive UI polish.
  • Reworked session persistence, save/restore, reconnect bootstrap, archived session recovery, session IDs, timestamps, and cross-day history restoration.
  • Added disk-backed chat logs, FRAME snapshots, chat-log search, generated session titles, the LOGS tab, session previews, and log deletion flows.
  • Added anonymous mode, model switching, configurable context windows, Brain-first execution, and runtime setting restoration.
  • Added context budgeting, overflow handling, cleanup warnings, tool-result pressure handling, and context-pressure visualization.
  • Reworked runtime storage and synchronization to use persisted runtime state instead of stale browser state as the source of truth.
  • Added and stabilized the Windows launcher, Python/runtime detection, one-click first-run setup, LM Studio autodetection, model selection, and first-run vision support.
  • Expanded regression coverage, cleaned and fixed the test suite, synchronized README/architecture/agent/runtime documentation, and refreshed screenshots and diagrams.

- persist delayed memory reports to files
- support multi-report append/remove
- remove runtime action boundary suppression
- clarify delayed memory save and append rules
- update runtime action regression coverage
- update bold text color
- add idle-driven extraction and two-phase
  long-term L4 memory consolidation
- add regression coverage
- add update_delayed_memory action for extending reports
- archive absorbed L4 facts from active brain context
- add pinned delayed-memory reports
- refresh README, theme assets, and regression coverage
- add default manual start commands
- clean up configuration table
- update ui with close and attach buttons
- update main screenshot
- persist L4 facts snapshot to memory/facts
- merge browser L4 stores into a single source of truth
- prevent empty/private-window snapshots from overwriting facts
- preserve merged fact provenance with source_fact_ids
- remove fixed side safe spacing from chat layout
- keep collapsed panels docked to screen edges
- carry previous chat reasoning into the next brain prompt outside follow-up ticks
- remove L4 confidence scoring
- log detailed L4 merge operations for create, update, reinforce, and ignore decisions
- harden stream validation against recurrent sentence loops and empty JIN_COLOR markers
- fix underscore emphasis rendering for unicode identifiers
- add regression coverage for memory prompts, L4 merges, stream loops, JIN_COLOR parsing, and formatter output
- render active and delayed memory as a rotating dashed segment with hover glow
- persist deleted fact tombstones and prevent stale snapshots from restoring facts
- restore deleted L4 facts directly from logger cards
- group extraction and merge request/response traces with structured skip diagnostics
- prune processed pending facts and block browser sync during active L4 updates
- highlight memory records and avatar rings referenced in JIN responses
- resolve delayed memory previews against the latest stored report state
- add an L4 facts ring to the runtime avatar
- improve pinned delayed memory indication
- fix for update delayed memory flow not detected by runtime
- add JIN_FOR_L4 marker for resolving meaningful long-term memory conflicts
- process focused clarifications through the L4 maintainer in the background without a follow-up
- preserve delayed memory references when related L4 facts are replaced
- validate marker payloads, lifecycle events, and replacement flows with tests
- update README.md
- fix response button visibility for empty/truncated L4 runs
- add better explanation for L4 truncation logs
- catch and log LM Studio API and stream errors
- add active bubble state while JIN writes an L4 update
- rename marker to UPDATE_L4_FACTS for clearer action intent
- extend reasoning citations across runtime, active and L4 memory
- keep citation highlights persistent and synced with panels and avatar rings
- fix for long term facts loop detection
- add anchor and absorbed fact handling for delayed memory
- migrate delayed memory actions to load/unload naming
- update delayed memory through partial JSON payloads
- clean delayed memory references when L4 facts are deleted
- replace delayed memory absorbed facts with facts_ids
- document live avatar behavior
- fix tests
- update live avatar refresh behavior
- cleanup default background image
- clean up L4 memory sync logging
- remove dubious rule for delayed memory save
- expose stream validator repeat thresholds through config
- add service usage fallback and focus brain context tab on submit
- add avatar memory layer visibility toggle
- tighten reasoning/action spacing and scene tint layering
- update client contract tests and cache-busting versions
- add delayed memory delete/restore sync
- keep loaded delayed facts visible in l4 context
- persist submitted answer rating visuals
- make active memory autonomous and hide reasoning fallback
- add delayed memory delete/restore and loaded-context highlighting
- rename delayed memory and skill flows to load/unload
- keep skill state visible in runtime context
- simplify reasoning loop validator output
- persist answer rating state and autonomous active memory
- append attached file content to the context max to 32k tokens
- add custom JIN bubble skin selector
- fix bootstrap from stale localStorage data
- sync all JIN actions to same format
- cleanup old naming
- cleanup update_active_memory contract
- fix background websocket reconnect
- ui fixes
- add skill for agents public board
- add dynamic L-T lazy loading based on context window budget
- tighten agents board skill flow
- add L-T facts delete from delayed memory dropdown
- add model load and prompt progress indication
- update delayed memory payload lt facts naming
- update config.example.py
- update schema.jpg
- fix previous chat bootstrap follow-up message
- fix cross-origin requests
- fix background process copy on reload
- fix lt store concurrency
- fix context blocks ordering
- fix tests
- update Posting Board skill actions info
- fix inline code rendering
- disable new tabs on load when session is active
- fix posting board tool reuse and inbox instruction
- fix save delayed memory code quote
- fix update L-T race
- add context pressure rays indication
- sync memory storage with local files
- sync project docs
- ui fixes
- update JIN launcher
- update context overflow follow-up flow
- update clean up tool results contract
- add L-T deduplication request
- add MCP client, CALL_MCP action, and Blender skill
- simplify context blocks and remove waiting telemetry
- move file and project actions into dedicated skills
- update action contracts, file timestamps, tests, and docs
- split actions dispatcher into separate files
- refactor actions flow
- fix marker handle inside payload
- remove action counters
- ui fixes
- remove redundant telemetry from console
- fixes and updates for blender_mcp flow
- ui fixes
- sync runtime action bare fallback and ownership
- remove visual action sequence
- fix clean tool results turn boundary
- cleanup after save active memory refactor
- remove runtime stream guard duplication
- cleanup brain non-stream path
- unify l-t fact semantic signature
- fix stream malformed aliases
- cleanup dead code
- sync project documentation
- fix tests
- add first-run one-click default setup
- fix first run bootstrap
- add lm-studio auto detection
- update README.md
- update context modal ui
- add logs tab for memory panel
- add session title
- fix and cleanup tests
- sync documentation with latest changes
- sync documentation with latest changes
- add first-run vision support
- update screenshots
@makeitdouble makeitdouble self-assigned this Oct 1, 2026
@makeitdouble
makeitdouble merged commit 8ad500a into master Oct 1, 2026
1 check passed
@makeitdouble
makeitdouble deleted the feature/L4_memory_layer branch October 1, 2026 16:22
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant