2026-03-19 - 2026-06-19
Overview
94 Issues closed from 1 user
Closed
#121 integrate stars
Closed
#120 integrate TAROT
Closed
#94 feat: feature-to-prompt pipeline — batch context materialization via Airflow
Closed
#96 research: LLM fine-tuning pipeline — tip reactions as training signal
Closed
#83 research: next-gen ranking policies — Thompson sampling, neural bandits, hybrid
Closed
#128 chore(scheduler): skip agents whose data sources aren't granted
Closed
#129 feat(clustering): LLM-enrichment before embedding for better semantic clustering
Closed
#123 tip clustering does not work. need to improve an agent that semantic cluaterization of tasks from taskpile
Closed
#122 when i press “snooze” the pipeline seems broken
Closed
#127 refactor(consents): data-source consents only; drop per-agent consent gate
Closed
#113 agents/focus-area: auto-infer focus areas from task clustering
Closed
#97 feat: embedding-based task clustering — nomic-embed-text for semantic dedup + pattern extraction
Closed
#112 agents/time-of-day: auto-infer quiet hours, peak hours, timezone drift
Closed
#116 agents/recent-patterns: adaptive lookback window + weekly cycle detection
Closed
#114 agents/momentum: auto-infer per-user momentum baseline
Closed
#115 agents/overdue-task: auto-infer per-user lateness tolerance
Closed
#110 no any docs found
Closed
#109 why there is a “running” simulation?
Closed
#107 what’s the point of Ops section in admin panel?
Closed
#101 notify me button in the main user interface must be moved to the config page, where all the config and connections to external services go
Closed
#100 user feedback should be only “done”, “snoose” or “dismiss”, no “useful/not useful”
Closed
#102 a small button with gears icon should be plance on the main interface with the link to the config page of a user
Closed
#106 make airflow up, why it is down?
Closed
#95 feat: automated prompt optimization loop — sim A/B → promote winner
Closed
#93 research: model benchmark for tip generation — qwen2.5 vs llama3.2 vs gemma3
Closed
#103 user table in admin should show user id
Closed
#104 links to mlflow and airflow do not work
Closed
#105 token admin authentication
Closed
#18 Observability baseline: structured logs, Sentry, trace IDs across services
Closed
#99 feat: bandit consumes profile features (egreedy-v2, D=12)
Closed
#78 feat: signal source abstraction — generalize beyond Todoist
Closed
#98 ml/serving JetStream consumer for signals.> + feedback.>
Closed
#61 Per-user freshness SLAs for features
Closed
#54 Event schema registry + protobuf CI gate
Closed
#81 feat: feature registry + user profile builder
Closed
#84 research: LLM prompt strategies for tip generation quality
Closed
#92 feat: LLM tip quality monitoring dashboard in admin
Closed
#21 events: NATS JetStream ingress + normalized event schema
Closed
#22 Todoist sync via events (not on-demand fetch)
Closed
#88 feat: context assembler — user signals → structured LLM prompt context
Closed
#91 feat: prompt versioning — track prompt_version + model in tip_scores
Closed
#90 feat: LLM output validation + structured JSON retry for tip generation
Closed
#86 infra: ai compose profile — Ollama + LiteLLM for local dev
Closed
#80 refactor: generalize recommendation pipeline — candidate → rank → render stages
Closed
#89 feat: TipCandidate shared schema — typed candidates through the bandit pipeline
Closed
#87 feat: AI gateway — wire ml/serving to LiteLLM with model aliases
Closed
#79 feat: AI tip generation via Ollama — contextual advice from user signals
Closed
#82 feat: tip kind system — task, advice, insight, reminder
Closed
#77 bug: port mismatch — config.ts defaults to 3078, docker-compose uses 3001
Closed
#76 bug: no data retention purge for tip_scores and tip_feedback (GDPR 30-day)
Closed
#74 bug: Todoist access tokens never refreshed — silent failure after expiration
Closed
#75 bug: reward updates to ml/serving silently swallowed on failure
Closed
#73 bug: TipFeedback type includes helpful/not_helpful but backend rejects them
Closed
#57 Bandit v1: global-then-personalize LinUCB
Closed
#2 docker-compose dev environment
Closed
#10 integrations: Todoist connector (OAuth2, read tasks, complete task)
Closed
#52 Tip-kind-aware UI affordances
Closed
#11 recommender: service skeleton with POST /recommend and policy registry
Closed
#9 integrations: Connector interface + encrypted token vault
Closed
#1 Monorepo tooling: pnpm workspaces + uv/poetry + turbo + pre-commit
Closed
#53 Connector interface: revoke() is mandatory; surface provider-side revocation
Closed
#5 Secrets convention and .env.example per service
Closed
#58 ML observability baseline (alongside first model)
Closed
#4 Shared types package: OpenAPI source of truth
Closed
#6 auth: Google OAuth sign-in
Closed
#8 profile: minimal User model + first-login creation
Closed
#56 Shadow-deploy infrastructure for policies
Closed
#71 admin: read-only SQL runner against Postgres replica (sunsets to Superset in M4)
Closed
#70 admin: ops actions — replay signal, disable/promote policy, force-retrain, revoke token
Closed
#72 admin: health rollup — gateway, ml/serving, DB, event bus, MLflow, Grafana status
Closed
#67 admin: recommendation log — per-tip explainability (features, policy, score, feedback)
Closed
#65 admin: model registry panel — embed MLflow at /admin/models
Closed
#69 admin: data quality widget — missing-signal rate, stale-token rate, feature NaN heatmap
Closed
#66 admin: experiment dashboard — LinUCB per-arm stats and bandit reset
Closed
#68 admin: reward analytics — reaction distribution over time, per-policy compare
Closed
#64 admin: feature store browser — features sent to ml/serving per scoring call
Closed
#63 admin: event stream viewer — live tail of signals.* events
Closed
#62 In-app feedback: one-tap 'was this tip helpful?'
Closed
#49 Privacy: ToS/PP docs, consent capture, versioned per-user acceptance hash
Closed
#48 Compose profiles: core and full
Closed
#51 Metrics baseline wired into the prototype
Closed
#59 License decision + LICENSE file
Closed
#55 Web Push (VAPID) — deliver the moment, not the app
Closed
#50 Account deletion endpoint (DELETE /me) — Phase 0
Closed
#26 Feedback loop: reactions → rewards
Closed
#23 Feature store skeleton with five v1 features
Closed
#12 recommender: RandomPolicy v0 — pick one Todoist task uniformly
Closed
#19 Deploy Phase 0 to a single VM via docker-compose + Caddy
Closed
#24 ml/serving: FastAPI scorer + RemotePolicy in recommender
Closed
#16 web: tip page (the product)
Closed
#15 web: connect page
Closed
#14 web: sign-in page
Closed
#13 gateway: BFF with auth check and service fan-out
Closed
#25 ContextualBanditPolicy (LinUCB) replacing Random
128 Issues created by 1 user
Opened
#9 integrations: Connector interface + encrypted token vault
Opened
#4 Shared types package: OpenAPI source of truth
Opened
#12 recommender: RandomPolicy v0 — pick one Todoist task uniformly
Opened
#16 web: tip page (the product)
Opened
#15 web: connect page
Opened
#5 Secrets convention and .env.example per service
Opened
#3 Gitea Actions CI skeleton
Opened
#7 auth: Apple OAuth sign-in
Opened
#6 auth: Google OAuth sign-in
Opened
#14 web: sign-in page
Opened
#8 profile: minimal User model + first-login creation
Opened
#10 integrations: Todoist connector (OAuth2, read tasks, complete task)
Opened
#11 recommender: service skeleton with POST /recommend and policy registry
Opened
#2 docker-compose dev environment
Opened
#1 Monorepo tooling: pnpm workspaces + uv/poetry + turbo + pre-commit
Opened
#13 gateway: BFF with auth check and service fan-out
Opened
#29 integrations: generic webhook ingress
Opened
#35 notifier: push/email delivery with quiet hours + per-channel rate limits
Opened
#33 iOS app (SwiftUI) — sign-in, connect, tip
Opened
#28 integrations: Apple Health (web import) connector
Opened
#21 events: NATS JetStream ingress + normalized event schema
Opened
#27 integrations: Google Calendar connector
Opened
#30 profile: unified Profile model (prefs, contexts, consents)
Opened
#31 Client-side timing signals: location, idle, focus windows
Opened
#32 Advice library + mixing policy
Opened
#25 ContextualBanditPolicy (LinUCB) replacing Random
Opened
#26 Feedback loop: reactions → rewards
Opened
#22 Todoist sync via events (not on-demand fetch)
Opened
#20 Phase 0 E2E: Playwright suite covering sign-in → connect Todoist → see a tip
Opened
#17 web: offline caching + reaction queue
Opened
#18 Observability baseline: structured logs, Sentry, trace IDs across services
Opened
#34 Android app (Compose) — sign-in, connect, tip
Opened
#19 Deploy Phase 0 to a single VM via docker-compose + Caddy
Opened
#23 Feature store skeleton with five v1 features
Opened
#24 ml/serving: FastAPI scorer + RemotePolicy in recommender
Opened
#44 Multi-region failover + PITR backups
Opened
#46 Billing + subscription tiers
Opened
#45 Public integration SDK
Opened
#39 Online experiments framework: A/B + multi-armed bandits as first-class
Opened
#40 Cohort analysis + opt-in collaborative features
Opened
#38 MLflow model registry + shadow deploys
Opened
#41 Drift monitoring + model cards
Opened
#36 Decide-and-deliver scheduler
Opened
#42 Audit logging + token rotation + SOC2-lite controls
Opened
#43 Kubernetes deploy + horizontal autoscaling
Opened
#37 Orchestrated retrains (Prefect or Airflow)
Opened
#49 Privacy: ToS/PP docs, consent capture, versioned per-user acceptance hash
Opened
#48 Compose profiles: core and full
Opened
#47 Modular-monolith packaging + import-boundary lint
Opened
#50 Account deletion endpoint (DELETE /me) — Phase 0
Opened
#53 Connector interface: revoke() is mandatory; surface provider-side revocation
Opened
#51 Metrics baseline wired into the prototype
Opened
#52 Tip-kind-aware UI affordances
Opened
#57 Bandit v1: global-then-personalize LinUCB
Opened
#59 License decision + LICENSE file
Opened
#62 In-app feedback: one-tap 'was this tip helpful?'
Opened
#60 k3s intermediate stop on the path to full k8s
Opened
#56 Shadow-deploy infrastructure for policies
Opened
#58 ML observability baseline (alongside first model)
Opened
#61 Per-user freshness SLAs for features
Opened
#55 Web Push (VAPID) — deliver the moment, not the app
Opened
#54 Event schema registry + protobuf CI gate
Opened
#63 admin: event stream viewer — live tail of signals.* events
Opened
#64 admin: feature store browser — features sent to ml/serving per scoring call
Opened
#65 admin: model registry panel — embed MLflow at /admin/models
Opened
#70 admin: ops actions — replay signal, disable/promote policy, force-retrain, revoke token
Opened
#69 admin: data quality widget — missing-signal rate, stale-token rate, feature NaN heatmap
Opened
#68 admin: reward analytics — reaction distribution over time, per-policy compare
Opened
#71 admin: read-only SQL runner against Postgres replica (sunsets to Superset in M4)
Opened
#67 admin: recommendation log — per-tip explainability (features, policy, score, feedback)
Opened
#66 admin: experiment dashboard — LinUCB per-arm stats and bandit reset
Opened
#72 admin: health rollup — gateway, ml/serving, DB, event bus, MLflow, Grafana status
Opened
#77 bug: port mismatch — config.ts defaults to 3078, docker-compose uses 3001
Opened
#73 bug: TipFeedback type includes helpful/not_helpful but backend rejects them
Opened
#75 bug: reward updates to ml/serving silently swallowed on failure
Opened
#74 bug: Todoist access tokens never refreshed — silent failure after expiration
Opened
#76 bug: no data retention purge for tip_scores and tip_feedback (GDPR 30-day)
Opened
#80 refactor: generalize recommendation pipeline — candidate → rank → render stages
Opened
#78 feat: signal source abstraction — generalize beyond Todoist
Opened
#79 feat: AI tip generation via Ollama — contextual advice from user signals
Opened
#81 feat: feature registry + user profile builder
Opened
#82 feat: tip kind system — task, advice, insight, reminder
Opened
#84 research: LLM prompt strategies for tip generation quality
Opened
#83 research: next-gen ranking policies — Thompson sampling, neural bandits, hybrid
Opened
#85 Consolidate MLflow + Airflow auth into shared OIDC provider
Opened
#86 infra: ai compose profile — Ollama + LiteLLM for local dev
Opened
#87 feat: AI gateway — wire ml/serving to LiteLLM with model aliases
Opened
#88 feat: context assembler — user signals → structured LLM prompt context
Opened
#89 feat: TipCandidate shared schema — typed candidates through the bandit pipeline
Opened
#90 feat: LLM output validation + structured JSON retry for tip generation
Opened
#91 feat: prompt versioning — track prompt_version + model in tip_scores
Opened
#92 feat: LLM tip quality monitoring dashboard in admin
Opened
#93 research: model benchmark for tip generation — qwen2.5 vs llama3.2 vs gemma3
Opened
#94 feat: feature-to-prompt pipeline — batch context materialization via Airflow
Opened
#95 feat: automated prompt optimization loop — sim A/B → promote winner
Opened
#96 research: LLM fine-tuning pipeline — tip reactions as training signal
Opened
#97 feat: embedding-based task clustering — nomic-embed-text for semantic dedup + pattern extraction
Opened
#98 ml/serving JetStream consumer for signals.> + feedback.>
Opened
#99 feat: bandit consumes profile features (egreedy-v2, D=12)
Opened
#100 user feedback should be only “done”, “snoose” or “dismiss”, no “useful/not useful”
Opened
#101 notify me button in the main user interface must be moved to the config page, where all the config and connections to external services go
Opened
#102 a small button with gears icon should be plance on the main interface with the link to the config page of a user
Opened
#103 user table in admin should show user id
Opened
#104 links to mlflow and airflow do not work
Opened
#105 token admin authentication
Opened
#106 make airflow up, why it is down?
Opened
#107 what’s the point of Ops section in admin panel?
Opened
#109 why there is a “running” simulation?
Opened
#110 no any docs found
Opened
#111 meta: shared context-inference framework for agents
Opened
#112 agents/time-of-day: auto-infer quiet hours, peak hours, timezone drift
Opened
#114 agents/momentum: auto-infer per-user momentum baseline
Opened
#116 agents/recent-patterns: adaptive lookback window + weekly cycle detection
Opened
#115 agents/overdue-task: auto-infer per-user lateness tolerance
Opened
#113 agents/focus-area: auto-infer focus areas from task clustering
Opened
#117 random policy should return random tasks, not todoist tasks
Opened
#118 Remove MLflow tracing from production deployment
Opened
#119 integrate google health
Opened
#120 integrate TAROT
Opened
#121 integrate stars
Opened
#122 when i press “snooze” the pipeline seems broken
Opened
#123 tip clustering does not work. need to improve an agent that semantic cluaterization of tasks from taskpile
Opened
#124 need to add openclaw and hermes memory systesm
Opened
#125 add the “journal” feature
Opened
#126 finish the mechanics of active notifications of the forgotten tasks
Opened
#127 refactor(consents): data-source consents only; drop per-agent consent gate
Opened
#128 chore(scheduler): skip agents whose data sources aren't granted
Opened
#129 feat(clustering): LLM-enrichment before embedding for better semantic clustering