Changelog

Release notes for FlowMail. Updates are delivered automatically through the app.

v0.2.28 (2026-04-06)

• Overlap calendar events with transparency, move agent button to top bar, filter sent emails from triage

v0.2.27 (2026-04-05)

• Add agent drawer panel accessible from any view via Cmd+J
• Rename Planning → Agent in Learning tab to match backend domain migration
• Fix 8 UI issues: session delete, calendar recurring events, agent error recovery

v0.2.26 (2026-04-05)

• Cache filtered tools instead of rebuilding on every agent loop iteration
• Handle model_switched event in agentStore
• Render model_switched events in agent transcript
• Implement model switching in agent loop (think_deeper/think_faster)
• Add think_deeper and think_faster tool definitions
• Add ModelSwitched event variant for agent model switching
• Add agent model switching implementation plan
• Add agent model switching (think_deeper/think_faster) design spec
• Add agent session transcript interactivity, event persistence, and tool state notifications
• Show SR badge for self-reflection sessions in Plans page

v0.2.25 (2026-04-05)

• Fix event data parsing bug in SelfReflectionTab, remove unused var
• Wire self-reflection auto-trigger, add Self-Reflection tab to Learning page
• Add agent session plan linking implementation plan
• Simplify: batch positive signals, fix fail-open guard, remove dead code, fix stringly-typed domain
• Merge feature/self-reflection-sessions: personality-driven self-reflection with restricted toolset
• Add self-reflection session trigger with personality-scaled threshold
• Add self-reflection tools: trigger_l1, propose/retire LEAP rules, report
• Add self-reflection prompt template with personality context
• Add V058 migration: reviewed_at column for self-reflection tracking
• Merge feature/positive-signals-auto-l1: positive signal tracking, regression detection, auto L1 triggering
• Record triage positive signals on sync, auto-trigger L1 on degradation
• Record positive signals for draft acceptance and plan completion
• Add signals module for positive signal recording and regression detection
• Add success tracking fields and query functions for positive signals
• Add V057 migration: success_count for positive signal tracking
• Make L1 apply_component auto-execute (no approval checkpoint)
• Add Phase C implementation plan: self-reflection sessions
• Add Phase B implementation plan: positive signals and auto L1 triggering
• Merge feature/unified-learning-pipeline: unify 6 feedback domains to 3, personality-aware agent reflexions
• Remove dead PlanReflexionInput, replace with lean PlanCorrectionCtx
• Simplify: extract memory context helper, fix stringly-typed domain, update dead match arms
• Remove old reflexion templates, keep triage + agent only
• Reroute task and memory corrections to agent reflexion domain
• Reroute plan corrections to agent reflexion domain
• Enrich agent tool rejection with email/card/tool context
• Add spawn_agent_reflexion with personality-aware LLM reflexion generation
• Add AgentReflexionInput and build_agent_reflexion_prompts
• Add reflexion_agent.toml prompt template with personality context
• Simplify FeedbackDomain: 6 variants to 3 (Triage, Draft, Agent)
• Add V056 migration: consolidate feedback domains to agent
• Add Phase A implementation plan: unified learning pipeline foundation
• Add unified learning pipeline design spec
• Fix Agent tab: session→transcript, plan→DAG, add New Session button
• Simplify: extract autoGrow util, use shared Button, remove comment noise
• Add Personality tab to Settings page
• Add CommunicationSection component with tone, sliders, and text fields
• Add IdentitySection component with auto-growing text areas
• Add DrivesSection component with labeled personality sliders
• Add personalityStore for frontend personality state management
• Wire autonomy and self_improvement_urgency drives to thresholds
• Inject personality config into agent system prompt
• Add PersonalityConfig struct and prompt renderer
• Add Phase 1 implementation plan for agent personality system
• Add agent personality system design spec

v0.2.24 (2026-04-05)

• Simplify: fix session flash bug, extract selectedSession derived
• Wire AgentTranscript into session detail view alongside plan DAG
• Add AgentTranscript component for historical session event display
• Make session rows clickable, load linked plan into DAG view
• Simplify: fix UUID ordering bug, add plans.agent_session_id index, remove dead list_sessions
• Show linked plan info in agent session list
• Triage wraps plan generation in agent sessions
• CreatePlanTool links plan to session and auto-executes
• Simplify: fix service import, card_id nullability, extract finalize_session helper, clean up stubs
• Dispatch creates agent session instead of plan, navigates to Agent tab
• Add agent session list to Agent tab
• Rename Plans tab to Agent across sidebar, command palette, and page header
• Add frontend service functions for agent session queries
• Add Tauri commands for querying agent sessions and events
• Persist agent events to DB during agent loop execution
• Add agent session and event query functions with tests
• Add agent_sessions and agent_events tables (migration V054)
• Add Phase 1 implementation plan: agent session persistence
• Update spec: agent sessions as primary unit, Plans tab → Agent tab, dispatch integration
• Add agent-orchestrated execution design spec (3 phases)

v0.2.23 (2026-04-04)

• Migrate remaining 5 tools to unified registry, clean up dead handlers
• Simplify: fix bugs and clean up from code review
• Add unified focus queue query for task-based approval surface
• Generate tasks from plan nodes requiring approval
• Add source_type/source_id to tasks for unified approval surface
• Migrate 28 tools to unified registry, clean up legacy fallbacks
• Wire agent loop to unified tool registry with legacy fallback
• Wire plan executor to unified tool registry with legacy fallback
• Add unified tool module with Tool trait and initial implementations
• Slim triage classification: 5 LLM fields, remove classify_email node, add read/action types

v0.2.22 (2026-04-04)

• Multi-intent triage classifier with structured plan handoff
• Fix get_pending_approval_plans missing analysis column, clarify Plans UI labels

v0.2.21 (2026-04-04)

• Fire-and-forget plan creation from Cmd+K agent dispatch
• Fix agent tool-use retry logic and system role handling
• Fix Gemma 4 runtime on newer Apple Silicon chips with Metal compilation and MoE fallback

v0.2.20 (2026-04-04)

• Add tool-use support for local AI models, enabling agent inference without cloud API calls
• Fix plan approval UI labels and missing analysis column

v0.2.19 (2026-04-03)

• Fix Gemma 4 model loading with multimodal builder

v0.2.18 (2026-04-03)

• Add Gemma 4 and Qwen 3.5 model families for local inference
• Switch to upstream mistralrs v0.8.0

v0.2.17 (2026-04-02)

• AI Agent system — Give natural language instructions via Cmd+K and the agent builds and executes plans on your behalf
• Multi-turn agent execution with a curated tool registry (search, draft, create plans, manage cards)
• Progressive trust: "Always allow" standing permissions for approved tool types
• Agent approval items appear in the Focus Queue for review before sensitive actions
• Agent execution transcript visible in the Plans page
• Streaming real-time output via SSE
• Context enrichment: agent receives email body, thread history, sender knowledge, and card instructions
• Automatic fallback from local to cloud LLM when needed

v0.2.16 (2026-04-01)

• Add web research plan nodes for gathering external information during plan execution
• Fix 7 UI bugs: calendar sync, settings panel, plans inspector, multi-select, reply-all, CC/BCC handling

v0.2.15 (2026-04-01)

• Calendar integration — Bi-directional sync with Gmail and Outlook calendars
• Three views: Month (with mini-calendar sidebar), Week (hour grid), and Agenda (list)
• Quick-create events by clicking time slots
• Card and plan integration: link events to cards, generate follow-up plans from events

v0.2.14 (2026-03-31)

• Plans page enhancements — Floating NodeInspector panel, 4-tab structure (Plans, Templates, Triggers, Catalog)
• Save completed plans as reusable templates with automatic template matching
• Scheduled triggers for automated recurring plan generation
• CreatePlanModal for building plans from natural language instructions
• Partial replan: regenerate remaining steps with user feedback while preserving completed work
• Plan analysis summaries in plan headers
• Meta-improvement button for triggering L2 reviews
• Calendar event plan follow-up
• Fix 5 Knowledge tab bugs: inbox buttons, graph filtering, type change, panel collapse

v0.2.13 (2026-03-31)

• Knowledge tab rewrite — Notes-app style UX with graph visualization and inline editing

v0.2.12 (2026-03-31)

• Fix migration crash: disable FK constraints during migration to prevent V045 FK violation on upgrade

v0.2.11 (2026-03-31)

• Fix startup crash caused by V045 migration foreign key violation

v0.2.10 (2026-03-30)

• Knowledge graph — New unified system replacing the old Memory and People views
• Entity types: person, organization, project, concept, fact
• Interactive graph visualization with node types and edge relationships
• Health scoring (confidence, recency, access frequency) for all entities
• Vector embeddings and semantic search across your knowledge base
• Entity inbox for reviewing automatically extracted entities before saving
• Connection panels showing how entities relate to each other
• Rumination system — Background self-play process that generates diverse classification hypotheses to improve AI accuracy
• Remove 1,210 lines of dead contact/memory code

v0.2.9 (2026-03-30)

• Fix memory leaks, draft collapse, markdown preview, Cmd+K behavior, navigation order, and sent tab display

v0.2.8 (2026-03-30)

• Fix plan canvas: menu overflow, node/edge deletion, button styling

v0.2.7 (2026-03-30)

• Add node kind catalog with 23 node types
• Add calendar and delegation plan handlers

v0.2.6 (2026-03-30)

• Improve plan approval gate behavior

v0.2.5 (2026-03-30)

• Add plan approval gate: PendingApproval status with rendered prompt viewing

v0.2.4 (2026-03-29)

• Add plan feedback system: correction snapshots, node annotations, partial replan feedback, DAG-level feedback bar

v0.2.3 (2026-03-29)

• Add inline prompt viewer to Components tab, remove dead model selectors
• Add implementation plan: benchmark Phase 3 - pipeline benchmarks
• Add memory task type to benchmark UI: config, results, test case generation
• Add memory benchmark: benchmark_memory, run_memory_test, score_memory, test case generation
• Fix plan executor skipping Ready nodes and graph over-zoom
• Add memory benchmark fields: fact_precision, fact_recall, entity_accuracy
• Add implementation plan: benchmark Phase 2 - memory task type
• Add plan task type to benchmark UI: config, results, test case generation
• Add plan benchmark: benchmark_plan, run_plan_test, score_plan, test case generation
• Add plan benchmark fields: complexity_correct, urgency_delta, node_kinds_correct
• Add implementation plan: benchmark Phase 1 - plan task type
• Add design spec: comprehensive benchmark system update

v0.2.2 (2026-03-29)

• Fix: remove card_notes from feature-gated benchmark DraftContext

v0.2.0 (2026-03-29)

• Update Cargo.lock
• Fix review: FK constraint, transaction, compute mode, node delete, double load, click-outside
• Wire plan editing into PlansPage and add editable config to detail panel
• Add AddNodeMenu and enable interactive DAG editing in canvas
• Add plan mutation queries and commands: add/remove node/edge, update config
• Add Phase C full DAG editing implementation plan
• Implement PlansPage with split layout: plan list + DAG canvas
• Add list_plans query and command with status/level filters
• Add Plans to sidebar navigation with placeholder page
• Add Phase B Plans page implementation plan
• Fix review: cache plan data, null guard dagre, in-place selection, merge plan commands
• Integrate inline plan DAG canvas into Focus Queue TaskInputWidget
• Add plan inspector components: dagre layout, PlanNodeCard, DAGCanvas, DetailPanel
• Add get_plan_with_edges command returning plan + nodes + edges
• Add @xyflow/svelte and @dagrejs/dagre dependencies
• Add Phase A plan inspector implementation plan
• Add plan inspector and DAG editor design spec
• Fix review: correct prompt variables, batch planning count, use shared Button
• Add Components tab to Learning page showing improvable component registry
• Add Planning domain to Learning tab: reflexions, rules, prompts, thresholds
• Add deep tier inference settings to backend and frontend model store
• Add Deep variant to ModelTask enum to fix manifest deserialization
• Fix model defaults: Fast=Qwen3-4B, Standard=Qwen3-8B, Deep=Qwen3-32B
• Add three-tier model selector (Fast/Standard/Deep) with Qwen 3 defaults
• Add multi-input trigger commands and frontend polling for scheduled triggers
• Add V036 scheduled_triggers table with types and queries
• Add PlanEvent enum and generalize plan generator for multi-input triggers
• Add Phase 8 implementation plan: multi-input triggers
• Add trigger_meta_improvement command with configurable frequency via learning_settings
• Add L2 meta-improvement: plan generation, analysis handler, L1 proposal
• Add Phase 7 implementation plan: L2 meta-improvement
• Fix review: transaction for activate_version, log seed errors, fix node types and domain mapping
• Add Tauri commands for component registry and L1 improvement trigger
• Add L1 node handlers: retrieve_component, analyze, generate_improved, apply
• Add L1 improvement plan generation and improve_prompt template
• Add component seeding and DB-first prompt loading
• Add component registry: V035 migration, types, CRUD queries
• Add Phase 6 implementation plan: L1 improvement plans + component registry
• Fix review: log correction errors, truncate reflexion body, optimize node lookup
• Wire plan corrections into approve_plan_node and add frontend wrapper
• Add record_plan_correction command with background reflexion generation
• Add planning domain to reflexion system with prompt and correction queries
• Add Phase 5 implementation plan: planning corrections + LEAP
• Update planner prompt for extraction and add candidate review wrappers
• Implement extract_knowledge and update_contact plan node handlers
• Add Phase 4 implementation plan: knowledge extraction
• Fix review: unify node types, fix archive template, efficiency improvements
• Add template CRUD commands and seed default templates
• Integrate tiered generation: template match (Tier 0) before LLM (Tier 1)
• Add template matching, instantiation, and tier assignment
• Add plan template CRUD queries
• Add V034 plan_templates migration and PlanTemplate types
• Add Phase 3 implementation plan: tiered generation + plan templates
• Fix review issues: generator lock, NULL input_data, debounce events, json_extract, retry logging
• Add Tauri event listeners for real-time plan status updates
• Add wait condition support: query waiting nodes, resume on reply
• Rewrite executor for parallel DAG execution with Tauri event emission
• Refactor handlers to use Arc<Mutex<Connection>> with short lock holds
• Add Phase 2 implementation plan: parallel DAG execution
• Fix review issues: edge type, started_at, error handling, N+1 queries, From impl
• Add plan service and update Focus Queue to show plan nodes
• Add Tauri commands for plan operations: generate, focus queue, details, approve
• Add sequential plan executor with output passing and checkpoint support
• Add node handlers: classify, draft, send, archive, retrieve, extract
• Add Tier 1 light plan generator for email events
• Add plan_generator.toml prompt template and loader
• Add plan/node CRUD queries and Focus Queue query
• Add plan module with core types: Node, Plan, enums, FocusQueueItem
• Add V033 migration: plans, plan_nodes, plan_node_edges, plan_corrections
• Add Phase 1 implementation plan: core schema + linear plans
• Add FlowMail v2 plan-centric architecture design spec

v0.1.50 (2026-03-27)

• Skip triage for sent emails, extract memories, fix Sent Tab
• Fix triage keyboard navigation: highlight, scroll-into-view, and select-next after delete
• Fix email readability by overriding inline dark text colors in HTML emails
• Add Create Template option to template picker dropdown
• Add production context as benchmark parameter, fix charts and card accuracy
• Fix Focus Queue to use approve_and_send_draft for CIPHER learning
• Remove unused avatar placeholder from top bar

v0.1.49 (2026-03-27)

• Add benchmark comparison charts with metric selector and config toggles
• Fix clippy: use strip_suffix instead of manual suffix stripping

v0.1.48 (2026-03-27)

• Refactor BehaviorsTab to unified task type system
• Integrate template picker and save-to-template into compose components
• Add frontend template service, store, picker, and save modal
• Sync task types to V030 schema: add remember/ask, remove approval/input_needed
• Refactor behavior executor to unified create_task dispatch and fix triggers
• Add template Tauri commands with placeholder resolution and signature stripping
• Add reply_templates table, template_id on tasks, and behavior migration (V030)
• Fix benchmark results: add thinking column, richer detail, CSV/JSON export
• Fix email triage filter and add keyword detection to EmailPreview task input
• Add thinking mode toggle to benchmark config UI

v0.1.47 (2026-03-27)

• Add draft writing rules UI and opt-in Self-Refine for drafts
• Add max test cases limit to benchmark (default 5)
• Fix clippy: use is_some_and instead of map_or(false, ...)

v0.1.46 (2026-03-27)

• Fix all Svelte a11y warnings and unused CSS
• Remove broken Metal PagedAttention, fix benchmark to use production inference
• Add Reconnect button to email account settings

v0.1.45 (2026-03-27)

• Hide triaged emails from triage view by default
• Fix People tab: email direction, self-contact, org classification, auto-relationships
• Add account filtering in triage and account selection in focus compose
• Add docs link to Settings preferences
• Fix OAuth re-auth fallback: handle plain string errors from Tauri IPC

v0.1.44 (2026-03-27)

• Fix notify-site: use github.token to read release notes
• Allow clippy::too_many_arguments on contact update functions

v0.1.43 (2026-03-26)

• Fix calendar event creation: add timezone and non-blocking extraction
• Fix multi-account sync: restore incremental sync and accurate counts
• Connect task extraction to Reflexion+LEAP learning pipeline
• Remove Context panel (RightPanel) feature
• Fix CIPHER draft self-learning pipeline and learning tab display
• Add sent email view, account-specific signatures, and card drag fixes
• Persist new email drafts and add markdown preview in InlineCompose
• Add customizable keyboard shortcuts and fix People view navigation
• Add People page improvements: CRUD, sorting, org classification, email preview, network graph
• Fix thread-level importance reclassification in triage view
• Move releases to public repo and simplify updater

v0.1.42 (2026-03-26)

• Add domain filtering to Learning overview charts
• Add People system, delegate action, and keyword detection
• Move benchmark to Settings tab and update scoring to use action_needed
• Add Shift+Arrow keyboard multi-select in triage view
• Add schedule task type and scope Add to Calendar button to it