fix: PostgreSQL + TanStack Start migration (merge-ready)#9
Merged
Conversation
Replace text nav links with icon buttons (gear for settings). Convert new eval set page into a modal with sticky header/footer, backdrop blur, escape-to-close, and SSE abort on close. Remove settings breadcrumb. /sets/new now redirects to dashboard. Components: NewEvalSetContext (provider), Header, AppShell, NewEvalSetModal, NewEvalSetButton. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…rences Migrate workflow agents from internal /rest/api/v1/runworkflow to public /rest/api/v1/agents/runs/wait. Autonomous agents remain on /rest/api/v1/chat with full trace data. Workflow agents lose traces (accepted limitation — documented in docs/TRACE_API_LIMITATIONS.md). Remove internal docs (gko-eval-presentation, harness-engineering-plan, issues, glean-api-needs), unused @gleanwork/api-client dep, and all scio-prod references. Fix settings password field rendering and CaseEditor alert() calls. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
… Start and Node based Co-authored-by: Cursor <cursoragent@cursor.com>
Move the web app from Next.js/Bun to TanStack Start on React 19 with pnpm/Node runtime support. Preserve the Node CLI, shared SQLite/settings behavior, and migrate tests to Vitest. The prompt snapshot diff is format-only from Bun snapshots to Vitest snapshots; prompt contents were compared by snapshot name and are unchanged.
Remove dead modal create flow, move new-eval-set and eval-config into focused modules with React Query hooks, and fix set creation API.
Export AgentSchema from generate-agent and use it in CLI schema fetch paths so agent/set/generate commands share typed field shapes. Co-authored-by: Cursor <cursoragent@cursor.com>
Restore dev ui-refresh UX: gear-only header, global create-eval-set modal, icon dashboard CTA, /sets/new redirect, settings breadcrumb removal, and abort in-flight generation when the modal closes. Use native password input for the API key field.
Restore web token usage recording, make failed web runs surface as failed, and bring active web code under Biome coverage with the route/service cleanup integrated.
Preserve failed and running runs in latest-run displays so the TanStack UI does not hide partial failures behind completed-only stats.
Defines the architecture for Seer as a deployable, cloud-agnostic evaluation platform for Glean's solutions library. Covers repo structure, OAuth passthrough auth, data model, API contract, deployment model (Terraform), and 9-phase migration path. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Frame Seer around Glean-specific knowledge-work evals, human calibration, macro behavior patterns, and business outcome evidence.
Drop temporary TanStack migration planning artifacts now that the migration has landed.
Replace better-sqlite3/bun:sqlite with pg (node-postgres) across CLI and TanStack web app. Schema uses pgTable with typed jsonb columns via $type<T>(), eliminating ~25 manual JSON.stringify/parse calls. Bootstrap simplified from 180 lines of raw SQL to criteria seeding only — schema management moves to drizzle-kit push. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Use --env-file-if-exists so the CLI works locally with .env and in cloud deployments where DATABASE_URL is set by the platform. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Pre-merge review of dev surfaced several blockers in the SQLite→Postgres and Next.js→TanStack Start migrations: - CI: provision a Postgres service + DATABASE_URL + `db:push` step in check.yml so `pnpm test:web-api-smoke` (and any DB-touching job) can run. - DB bootstrap: initializeDB no longer swallows a missing-schema error and permanently marks the DB initialized; it now throws an actionable error and only marks initialized on success. - CLI: use withLedgerContext (AsyncLocalStorage) per case so --parallel runs attribute token usage to the correct case, and parseAsync + closeDB() so the open pg pool no longer keeps the event loop alive and hangs the CLI. - web: PATCH /api/cases returns 404 when the case ID does not exist. - smoke test: inject the partial-failure case via a sentinel query (empty agent response) instead of writing invalid JSON into the now-jsonb metadata column, which Postgres rejected at write time. Verified: typecheck, lint, unit tests (79), web build, and web-api-smoke (against a fresh Postgres) all pass. Co-authored-by: Cursor <cursoragent@cursor.com>
Route all outbound Glean requests through gleanFetch (undici) so SEER_GLEAN_TLS_INSECURE=1 bypasses cert verification in local dev. TLS/cert errors now fail fast (no retry) with actionable messages in both CLI and web UI. Agent lookup card shows errors inline and handles missing schemas gracefully. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add undici to web/package.json (was only in root, fragile hoisting) - Defer initializeDB() to preAction hook so --help/--version work without a running database - Update CLAUDE.md: SQLite → PostgreSQL (7 references) Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
PG migration, TanStack Start, TLS resilience. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Merge product vision and technical architecture into AGENTS.md. CLAUDE.md is now a symlink so both Claude and Codex read the same context. Updated architecture section for TanStack Start, PostgreSQL, gleanFetch, and current file layout. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
CI failed because pnpm-lock.yaml was missing the undici specifier for the web workspace package. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
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
2 participants
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
Promotes the
devmigration milestone tomain: Seer moves from SQLite → PostgreSQL and the web app from Next.js → TanStack Start, plus the customer-readiness work (unified Glean API key, internal references removed). This PR also includes the pre-merge review fixes that make the migration actually merge-ready.Two infra migrations land together:
data/seer.db) → PostgreSQL via Docker Compose + Drizzle (pg,jsonbcolumns,drizzle-kit push).web/src/routes/*,web/src/server/*, feature slices underweb/features/*). All 9 API routes reimplemented.Pre-merge review fixes (this branch's final commit)
A full review of
main...dev(155 files) surfaced blockers that are fixed here:check.ymlranpnpm test:web-api-smokewith no Postgres. Added apostgres:16-alpineservice,DATABASE_URL, and adb:pushstep.initializeDBswallowed a missing-schema error and permanently marked the DB initialized. It now throws an actionable error and only marks initialized on success.pgpool kept the event loop alive (CLI never exited on success);--parallelruns used shared ledger state. Now usesparseAsync+closeDB()andwithLedgerContextper case (matching the web run-service)./api/casesnow returns 404 for a missing ID instead of an empty 200.jsonbmetadata column (rejected by Postgres at write). Reworked to fail the case via a sentinel query that yields an empty agent response.Test plan
pnpm typecheck— passpnpm lint— passpnpm test— 79/79 passpnpm --filter web typecheck+build— passpnpm test:web-api-smoke— pass against a fresh Postgres (verified locally on a disposable container)Follow-up / open decision
data/seer.dbdata. Versioned migrations were removed in favor ofdrizzle-kit push. Greenfield is fine; anyone with an existing SQLite DB (local dev included) has no ETL path. Decide whether that data needs to be preserved before this is relied on.web/README.md,docs/architecture.md,CLAUDE.md).