Launch polish: project scoping, timeline landing, structured answers + SSE hardening#79
Merged
Conversation
- projects table (settings JSONB: default_timeline_window, auto_load_timeline,
max_events) + project_id on log_chunks/watcher_configs/conversations/
investigations; real signature column on log_chunks (un-defers Path B from
cluster_view) with idempotent backfill; Default project seed absorbs all
pre-project rows
- /projects CRUD + /projects/{id}/services; resolve_project shared name-or-id
resolver (name get-or-create, uuid must exist, blank -> Default)
- ingest stamps signature + project_id (API form field, worker via
watcher_configs.project_id); /watchers accepts project_id
- scoping: RetrievalFilters.project_id applies to both vector + FTS arms;
every ReAct tool gains project_id injected via container closures (LLM
never sees it; cache keys include it); chat + investigate accept/inherit
project_id; known_services resolved per project
- conversations list/detail return project_id + name for the sidebar
Verified live: zk ingested into 'Infra', ssh into Default; Infra-scoped chat
cites only Infra chunks; asking Infra about ssh-server clarifies with zero
cross-project leakage.
…, guided actions (UX P2)
- event_feed.py: deterministic event rules over per-(service, signature)
time buckets — begins / spike / subsides / new_pattern / health
transitions. Pure rule engine, no LLM cost per load; 15 unit tests.
- GET /projects/{id}/overview: events + corpus-wide signature clusters
(the un-deferred Path B) + services + derived suggested actions.
Window anchors to now and falls back to the project's latest data so
the landing page always tells the most recent story available.
- UI: ProjectPicker (0 projects -> create, 1 -> auto-select, 2+ -> cards)
replaces the empty-chat hero; ProjectOverview renders the timeline,
clusters, services and suggested-action chips as the landing panel;
chips route to Deep Research (grounded query: signature + service +
time range pre-filled) or /chat; sidebar shows per-conversation
project badges; chat + investigate carry the conversation's project.
Verified live on real LogHub data: Infra overview shows 'zookeeper enters
degraded state -> recovers' with clusters and 5 action chips; clicking an
Investigate chip ran a scoped DR investigation end-to-end. 6/6 puppeteer
flows pass against the new flow.
- parse Apache error-log format ([Sun Dec 04 ...] [error] ...) with level mapping - tag syslog pam/sshd auth-failure bodies WARNING so error-scans surface them - signature masking: collapse IPv4 to <IP>; preserve HTTP status codes, protocol versions, and mid-identifier digits (jk2_init) — high-cardinality tokens still mask - sources.md records public datasets used for real-log testing
- _compact_observation clips tool results fed back to the LLM (lists to 10 items, text to 300 chars, valid-JSON fallbacks) — full results still persisted to DB/ledger; sweep context no longer pretty-printed - shared _post_with_429_retry: Retry-After-aware 429 waits for OpenAI, Anthropic, Gemini (previously Mistral-only) - loop retries honor retry_after and never retry LLMBadRequestError
Steps only stream after a tool observation completes, leaving 10-30s of dead air. ThinkingIndicator fills the gap with a contextual status line (per-tool wording, reflection, compiling) rotating through generic thinking words every 3s. Hidden on done/error/clarification.
The compiled InvestigationAnswer is persisted as json.dumps(...), but the UI rendered it as plain text — so the final card showed raw JSON. Add a CompiledAnswer component that parses the answer and renders root cause, confidence, affected services, trigger, propagation chain, ruled-out hypotheses, assumptions and gaps; non-JSON answers (clarification text, legacy prose) fall back to plain text. SSE robustness: - Live clarifications now emit `clarification_request` instead of a meaningless `done` with "Awaiting clarification..." — previously the question was lost until a page reload hit the replay path. - json.dumps in the stream generator uses default=str so a stray non-serializable observation can't kill the stream. - Client drops a single unparseable frame instead of tearing down the connection, and tolerates EventSource auto-reconnect on transient disconnects rather than surfacing a hard error.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
VarunGitGood
added a commit
that referenced
this pull request
Jun 14, 2026
Launch polish: project scoping, timeline landing, structured answers + SSE hardening
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Launch-prep work bringing repi to v0.2.0 — UX redesign, ingestion/parser coverage, LLM efficiency, and final-answer / SSE fixes.
Highlights
UX redesign (P1/P2)
project_id.Investigation / answers
InvestigationAnswer.SSE hardening
clarification_requestinstead of a meaninglessdone(the question was previously lost until a page reload).json.dumpsin the stream usesdefault=strso a stray non-serializable observation can't kill the stream.Ingestion
/api/v1/) in signatures.LLM
Verification
pytest tests/→ 254 passednpx tsc --noEmit+npm run build→ cleandocker compose build app→ image builds clean;docker compose configvalidLaunch note
The structured-answer fix ships in the web bundle, so the published image needs a rebuild + push to deliver it.