Repository navigation
feat: migrate to tanstack start and add all dashboard pages - #1
Merged
Merged
Conversation
steebchen
approved these changes
Apr 13, 2025
9 tasks
smakosh
added a commit
that referenced
this pull request
May 8, 2026
## Summary - Adds a public `/apps` page that lists tools and coding agents using LLM Gateway, ranked by tokens processed (similar to openrouter.ai/apps) - Aggregation comes from a new public endpoint `GET /public/apps` that groups `log.source` and sums `total_tokens` - Seed data expanded with popular coding agents — Claude Code, Cursor, Cline, Codex, OpenCode, Aider, Continue, Windsurf, Roo, Kilo, Zed, Bolt, v0, Lovable, Autohand, SoulForge, OpenClaw, n8n — and log volume bumped so the leaderboard has realistic ranks - Page features a top-3 podium with blue-accented #1, a top-12 grid, a table-style long-tail list, search + category filters, and a DevPass upsell as the closing section ## Design notes Aesthetic intentionally matches the existing landing-page DNA in `apps/ui/src/components/landing/*`: `font-display` headlines, `AnimatedGroup blur-slide` reveals, atmospheric blue radial blobs, gradient hairline separators, single `ShimmerButton` primary CTA. No purple-gradient AI-template tropes. The DevPass upsell uses logo-stack social proof from the leaderboard above (Claude Code, Cursor, Cline, OpenCode) and loss-aversion framing ("Stop juggling nine separate subscriptions") rather than generic feature bullets. ## Test plan - [ ] `pnpm run setup` to reseed; visit `http://localhost:3002/apps` - [ ] Verify the top 3 podium shows the highest-volume seeded agents - [ ] Verify the search box filters by display name and source slug - [ ] Verify the category pills (All / Coding agents / Automation / Other) filter correctly - [ ] Verify the long-tail list collapses below the top 12 - [ ] Verify `GET /public/apps` returns aggregated stats without auth - [ ] Verify `GET /public/apps?limit=5` respects the limit - [ ] Verify the DevPass upsell links to https://devpass.llmgateway.io - [ ] Verify dark mode looks correct (atmospheric glow, podium accent) 🤖 Generated with [Claude Code](https://claude.com/claude-code) <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Public "Apps" page with hero metrics (apps tracked, tokens processed, requests routed), ISR, and "Read the docs" link. * Searchable, category-filterable apps leaderboard with top-3 podium, results grid, and ranked list. * Public API endpoint providing aggregated app traffic metrics for the UI. * App metadata for richer app cards (names, descriptions, icons/links). * DevPass upsell hero with CTAs and example setup snippet. * **Chores** * Seed data expanded to generate richer, higher-volume demo activity for public metrics. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com> Co-authored-by: Luca Steeb (bot) <contact@luca-steeb.com>
This was referenced Jun 20, 2026
huangyingting
pushed a commit
to repomesh/llmgateway
that referenced
this pull request
Jun 28, 2026
## Why Diagnosed why the **chat plans** offer (Starter $9 / Plus $19 / Pro $49, shipped Jun 8) wasn't converting. Real data (PostHog, LLMGateway project): **7 conversions in ~3 weeks — 4 Plus, 3 Starter, 0 Pro** against ~106 monthly active playground users. Running the value equation, the binding constraint was **perceived likelihood**: the offer asked buyers to value abstract "credits / 2.5× multiplier" with no proof, no guarantee, and use-it-or-lose-it expiry framing. The funnel was also **blind** — chat pricing clicks and paywall impressions weren't tracked, so there was no way to see *where* it leaked. ## What changed **Make the value believable (the theopenco#1 lever)** - Translate each plan's credit allowance into concrete **"≈ N messages"** on frontier and fast models, e.g. Plus ≈ 3,000 frontier / 9,000 fast; Pro ≈ 9,300 / 28,000. New `estimateChatPlanMessages()` in `@llmgateway/shared`. - Estimates are **conservative**: anchored to the priciest model in each class (Claude Sonnet for frontier, Haiku for fast), so GPT-5 / Gemini / Flash all yield *more* messages than shown — never overstated. - Add a **"Replaces ~$60/mo of ChatGPT Plus + Claude Pro + Gemini"** anchor on frontier tiers; lead the page/paywall copy with the dream outcome instead of "credits." **Risk reversal + expiry reframe** - **7-day money-back guarantee** (refund if the plan is barely used — conditioned to avoid credit arbitrage, since credits are worth more than the price). - Reframe credit expiry from "don't roll over" → **"your allowance refills in full every cycle."** **Sharpen the tier ladder** - Starter copy is now **accurate**: it advertises the fast models it actually allows (Claude Sonnet, Haiku, Gemini Flash) and no longer claims GPT-5-mini, which is gated. Pro now has a tangible differentiator — the message-count jump over Plus. **Instrument the funnel** - Fire `pricing_plan_clicked` with `app: "chat"` on CTA clicks (mirrors the existing dev-plans event, so chat joins the same funnel). - New `chat_pricing_viewed` event on pricing-page and paywall views (`source` distinguishes them), making paywall impressions measurable for the first time. ## Out of scope (flagged for a product decision) - **Credit rollover** — reframed in copy, but the actual cycle-boundary mechanic is a billing change left untouched. - **Dropping / repricing tiers** — positioning only; no price or entitlement changes. - **Starter `gpt-5-mini` gating** — the blocked-pattern `"gpt-5"` also blocks `gpt-5-mini`/`gpt-5-nano` by substring. Copy is now accurate; whether to grant those cheap models to Starter is a product call. ## Testing - `turbo run build --filter=playground` ✅ (builds `@llmgateway/shared` first) - `pnpm format` ✅ · pre-commit lint-staged ✅ 🤖 Generated with [Claude Code](https://claude.com/claude-code) <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Refreshed pricing and paywall copy with clearer “one subscription” positioning for frontier models. * Added “≈ messages/mo” allowances and updated plan coverage wording. * Introduced a visible 7-day money-back guarantee. * **Analytics** * Enhanced pricing-page and plan interaction tracking for viewing and tier selection. * **Bug Fixes** * Updated “How it works” allowance language (fresh cycle/credit behavior) and revised plan details to better match the current offering. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
steebchen
added a commit
that referenced
this pull request
Aug 2, 2026
Stacked on #3234 (follows #3370, which merged into the same base). ## Problem The Provider Credentials page could not tell an operator that a credential had burned through its spend cap and was no longer eligible for selection: - A key the billing worker auto-disabled rendered as a plain `inactive` badge — indistinguishable from one an admin switched off by hand. - Every row was numbered by its position in the list, so a shut-off key read as `#1` in the rotation while the gateway was silently serving from the one below it. - There was no warning before a cap tripped, only after. - Org BYOK provider keys showed no spend information at all. ## Changes Spend-limit state is derived in one place (`ee/admin/src/lib/provider-key-spend.ts`) and shared by the managed-credentials page and the per-organization BYOK table: - **`limit reached` badge** no longer requires the status flip to have landed. Enforcement lags by one worker batch, and the key keeps serving traffic in that window, so an over-cap key that is still active renders as `limit reached · disabling` rather than a reassuring `active`. - **Usage bar** with an amber tint from 80% of the cap, so a key approaching its limit is visible before it trips. - **Rotation positions count only selectable keys.** Provider-key selection filters on `status = 'active'`, so out-of-rotation rows now show `—` with an explanatory tooltip and a de-emphasised background instead of a misleading rank. ## Verification - `pnpm format`, `pnpm build` — clean. - 62 tests pass across `provider-key-spend.spec.ts` (13 new, covering the state machine including the lag window and a zero/malformed cap), `provider-key-stats.spec.ts`, and `admin-provider-credentials.spec.ts`. - Driven in a real browser against the dev stack with all four states seeded (at-cap/inactive, 85% warning, over-cap/still-active, uncapped). Confirmed the badges, tooltips, usage bars and rotation numbering render as intended, and that the spend dialog reconciles: window total `$14.5125` matched the database, split across `FinTech Global $10.75` and `Test Organization $3.7625`. Both bucket granularities (hourly for 24h, daily for 7d) verified. Dev data restored afterwards. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: AYOUB BENDARSI <137005356+RATCHAW@users.noreply.github.com>
smakosh
added a commit
that referenced
this pull request
Aug 8, 2026
Converts the changelog entry into a blog post -- this is SEO/ marketing content, not a release note for a shipped feature. - new post at /blog/ai-gateway-benchmark, category Engineering, targeting "AI gateway benchmark"; adds an FAQ section and CTA block per the blog house style - OG image regenerated with the top-left reserved and the real logo-with-name-white.svg composited in, per the blog skill (gpt-image-2 cannot draw the mark correctly) - removes the changelog entry and its image Carries over the verified framing: leads with the #1 result, attributes it to tail latency rather than median speed, and keeps the caveats -- one run, tightly packed field, public history of lower placements. Still makes no causal claim for the week-over- week move, and the new FAQ states outright that a gateway adds a hop (our warm TTFT median is slower than the direct baseline). Claude-Session: https://claude.ai/code/session_018n5hdRdydty6rdPEhzPdrb
steebchen
added a commit
that referenced
this pull request
Aug 13, 2026
## Problem The [computesdk AI-gateway benchmark](https://github.com/computesdk/benchmarks/actions/runs/31716766465) (2026-08-13) dropped LLM Gateway from #1 (90.84, Aug 7) to #4 (88.80). The regression is entirely in cold-start probes: 3 of 10 stalled at 1.1–1.3s inside the gateway→provider hop while the remaining probes ran ~620ms, with edge DNS/TCP/TLS all under 10ms. A stall of that size is not a handshake. TCP+TLS to a provider's anycast edge is two round trips — measured at ~10ms against `api.anthropic.com` from a nearby vantage, and tens of ms at worst. 1.1s is a DNS retransmit: in-cluster an uncached lookup goes through `ndots:5` search expansion, where a single dropped UDP packet stalls for seconds. The dispatcher already caches DNS, but at a 30s TTL the entry expires between requests on a quiet pod, so an idle pod pays resolution again on its next request — exactly the cold-start case the probes measure. ## Approach - Default `UPSTREAM_DNS_CACHE_TTL_MS` 30s → 300s. Provider hostnames resolve to CDN/anycast addresses that are stable over minutes, and a connect failure on a stale address is already retried by provider fallback, so a long TTL is safe while a short one only puts DNS back on the TTFT path. - Plumb `upstreamDnsCacheTtlMs` and `upstreamKeepaliveTimeoutMs` through the Helm chart. Neither was settable without a code edit before; both stay unset in `values.yaml` so the code defaults apply, following the existing `terminationGracePeriodSeconds` pattern. This is the reduced form of #3595. That PR paired the TTL bump with a prewarm pinger that HEAD-pings provider origins to hold a pooled connection open. Measured against undici 8.9.0 with the same Agent config, every `HEAD` to both configured origins is answered `Connection: close`, so the ping opens a connection and has it torn down immediately — it cannot keep anything pooled: ``` HEAD https://api.openai.com/ 421 connection=close 6 connects / 3 req pooled=NO GET https://api.openai.com/ 421 connection=keep-alive 2 connects / 3 req pooled=NO HEAD https://api.openai.com/v1/models 401 connection=close 3 connects / 3 req pooled=NO GET https://api.openai.com/v1/models 401 connection=keep-alive 2 connects / 3 req pooled=NO HEAD https://api.anthropic.com/ 404 connection=close 3 connects / 3 req pooled=NO GET https://api.anthropic.com/ 404 connection=keep-alive 1 connect / 3 req pooled=YES HEAD https://api.anthropic.com/v1/messages 405 connection=close 3 connects / 3 req pooled=NO GET https://api.anthropic.com/v1/messages 405 connection=keep-alive 1 connect / 3 req pooled=YES ``` It is the method, not the path. Since the stall being chased is DNS rather than connection setup, the TTL change is the part that addresses it; a corrected pinger (`GET`, and preserving the configured path, which `new URL(x).origin` currently strips) can be revisited separately if connection setup turns out to matter. ## Verification - `pnpm vitest run apps/gateway/src/lib/upstream-dispatcher.spec.ts` — 3/3 passing - `helm template` with `gateway.config.upstreamDnsCacheTtlMs=600000` emits the var; with defaults it emits nothing - `pnpm build` — 17/17 tasks pass 🤖 Generated with [Claude Code](https://claude.com/claude-code) <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added configurable upstream DNS cache TTL and keep-alive timeout settings for gateway deployments. * Deployment configuration can now override the default upstream connection behavior. * **Improvements** * Increased the default upstream DNS cache duration from 30 seconds to 300 seconds. * DNS caching remains disabled when explicitly configured with a zero-second TTL. <!-- end of auto-generated comment: release notes by coderabbit.ai --> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
SeyedHashtag
pushed a commit
to SeyedHashtag/llmgateway
that referenced
this pull request
Aug 15, 2026
## Problem The July traffic report showed revenue concentrating in two places we under-serve: 1. **The DevPass dashboard is the theopenco#1 converting surface** (354 payers last month), and Reset Passes sold 42 units in month one with zero targeting — but the agreed follow-up from theopenco#3093 (a real cap-hit analytics event) never shipped, and the dashboard shows the same ResetPassCard at 0% and 100% usage. No contextual offer exists at the moment of highest intent. 2. **Compliance content converts payers at ~5%** (soc2-type-ii: 158 readers → 8 payers) but enterprise leads are flat at 4/month — the post never presents the enterprise path. Meanwhile the weekly report shows organic momentum fading, with the proven "[X] alternatives" and Kimi K3 playbooks sitting unshipped. ## What this ships ### DevPass: cap-hit funnel + contextual offer - **`devpass_premium_cap_rejected`** captured server-side in the gateway when the weekly premium-cap 402 fires (`assertDevPlanPremiumCapNotExceeded`), with `devPlan`, `model`, `msUntilReset`, and org group. This is the follow-up agreed in theopenco#3093: `devpass_weekly_cap_hit_viewed` only counts users who open the dashboard, but most cap hits happen inside coding agents that swallow the 402 — demand was undercounted. Adds a `posthog-node` client to `apps/gateway` (mirrors `apps/api/src/posthog.ts`; disabled without env, no hot-path cost). - **`CapHitResetOfferDialog`** on the DevPass dashboard: a visa-stamp dialog (border-control stamp, MRZ strip — house DevPass brand) that appears the moment the weekly premium cap is hit, driven by the existing 5s status poll. It mirrors the server's purchase/redeem gates (never offers an action the API would 400), stays quiet when the monthly pool is exhausted (that state belongs to `AllowanceExhaustedCard`), snoozes per cap-window via cookie, and emits `devpass_cap_hit_offer_shown/dismissed/clicked`. The CTA scrolls to the ResetPassCard rather than duplicating the purchase surface. - **`/ingest` PostHog proxy for `apps/code`** (mirroring apps/ui and apps/playground) — DevPass dashboard events were ad-blocker-droppable until now, so July's 500 cap-hit views were an undercount. Expect an event step-up after deploy. <img width="1080" alt="Cap-hit Reset Pass offer demo: dialog appears at 100% weekly usage, CTA scrolls to the ResetPassCard, redeem restores the allowance" src="https://raw.githubusercontent.com/theopenco/llmgateway/cc22b375431e6148e4889e403c4635327095fc0c/cap-hit-reset-pass-demo.gif" /> ([MP4 version](https://raw.githubusercontent.com/theopenco/llmgateway/cc22b375431e6148e4889e403c4635327095fc0c/cap-hit-reset-pass-demo.mp4)) ### Compliance/enterprise content cluster - `soc2-type-ii` gains a **provider compliance policies** section (the theopenco#3339 policy-aware picker, fail-closed requirements, 403-before-egress), a proper enterprise CTA block, and links into the new cluster. - Three sibling posts feeding the same funnel: **`llm-data-retention`**, **`gdpr-compliant-llm-routing`**, and **`llm-compliance-checklist`** — all fact-checked against `apps/docs/content/features/{data-retention,compliance}.mdx` and the routing docs (region example uses a real catalogue mapping, `aws-bedrock/claude-sonnet-4-6:eu-west-2`). - New **`BlogCta variant="enterprise"`** (→ `/enterprise#contact` + `/enterprise/compliance`) used by all three. ### Organic pipeline refill - **`portkey-alternatives`** and **`helicone-alternatives`** listicles — the two SERP gaps left open after the litellm/openrouter/copilot listicles proved the pattern. Facts per the verified June-2026 competitor landscape (Portkey→Palo Alto/Prisma AIRS; Helicone→Mintlify maintenance mode; Langfuse/LangSmith claims re-verified this week). Internal links added from `/compare/portkey`, the vs-Portkey post, best-ai-gateways, and both existing listicles' "skip" sections. - **Kimi K3 spokes**: `kimi-k3-open-weights` (weights shipped Jul 26 on HF under a custom **"Kimi K3 License"** — not the Modified MIT press predicted, so the post and pillar deliberately point at the LICENSE file instead of summarizing terms) and `kimi-k3-api` (the "kimi k3 api" query; reasoning_effort semantics, cached-input economics, sticky sessions). Pillar updated: weights-release facts corrected, spokes interlinked. - **`/rankings` interlinks** from the `/models` SEO copy and `llms.txt` (it had no entry there). - 7 OG images generated in the house circuit-board style with the composited wordmark. ## Verification - `pnpm format`, full `pnpm build`, and `dev-plans-reset-passes.spec.ts` (32/32) pass. - Demo recorded against the local stack with the real seed org: staged cap-hit via SQL (millisecond-truncated `dev_plan_premium_week_start` for the CAS), dialog fired, CTA scrolled to the card, redeem zeroed `devPlanPremiumCreditsUsed` and consumed the included pass server-side. ## Notes for review / follow-ups - The census dialog and the new cap-hit dialog can theoretically stack (both Radix dialogs; census mounts globally, cap-hit on the main dashboard page). In the wild both firing together should be rare; if we care, a simple priority gate in `DashboardShell` would fix it. - The premium-cap 402 still has no machine-readable `code` distinguishing it from out-of-credits — agents can't react programmatically. Left out deliberately (error-shape compatibility); worth its own PR. - Blog listicles ship FAQ sections but blog posts emit no FAQPage JSON-LD (model pages do). Separate SEO follow-up. - dev.to syndication for the new listicles is intentionally not part of this PR (staggering per the syndication policy). 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_018NeUff4XEsqAVuJRZ8KuvS <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit - **New Features** - Added a dashboard offer for eligible users who exhaust weekly premium usage, including reset-pass redemption and purchase options. - Added links to live model rankings and expanded enterprise compliance calls to action. - Added guidance for Kimi K3, GDPR-compliant routing, LLM compliance, data retention, and gateway alternatives. - **Documentation** - Updated model, compliance, licensing, and comparison content with new articles, refreshed links, and recommendations. - **Bug Fixes** - Improved analytics loading, routing reliability, and shutdown handling. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Co-authored-by: Luca Steeb <contact@luca-steeb.com>
SeyedHashtag
pushed a commit
to SeyedHashtag/llmgateway
that referenced
this pull request
Aug 15, 2026
> **Scope changed:** this started as a changelog entry and is now a blog post. Changelog is for shipped-feature release notes; this is SEO/marketing content about a third-party benchmark result, which belongs in `/blog`. The changelog entry and its image are removed in this PR. ## Problem We placed first in the August 7 run of [computesdk's independent AI gateway benchmark](https://www.computesdk.com/benchmarks/ai-gateway/) — composite 90.8, ahead of five other gateways and the no-gateway Anthropic control. Nothing communicates that to customers or prospects, and "AI gateway benchmark" is a keyword we don't currently rank for. ## Approach New post at `/blog/ai-gateway-benchmark` (category `Engineering`, id `blog-ai-gateway-benchmark`), plus its OG image. The post leads with the theopenco#1 result, then pivots to the axis that actually produced it — tail latency, not median speed: | Metric | LLM Gateway | Best of the rest | No-gateway baseline | | --------------- | ----------- | ---------------- | ------------------- | | Warm TTFT p95 | **1151 ms** | 1289 ms | 2035 ms | | Cold E2E p95 | **1262 ms** | 1266 ms | 1374 ms | | Cold E2E median | **594 ms** | 601 ms | 648 ms | That framing is deliberate: it's the enterprise-relevant story (predictability under load, which compounds in agentic workloads) and it's the one the data supports outright. Per the blog skill, the OG image was generated with the top-left reserved as negative space and the real `logo-with-name-white.svg` composited in afterward — gpt-image-2 hallucinates the mark if asked to draw it. ## Screenshots `/blog/ai-gateway-benchmark` at 1440px, rendered from the production build. <details open> <summary>Light</summary> <img width="1440" alt="Blog post 'Ranked theopenco#1 on an Independent AI Gateway Benchmark' in light theme, showing the OG image with composited logo, the composite leaderboard table and the tail-latency table" src="https://raw.githubusercontent.com/theopenco/llmgateway/5d0e1e7031cdafb8a4e06bf6f20a289ea5798f78/blog-light.png" /> </details> <details> <summary>Dark</summary> <img width="1440" alt="The same blog post in dark theme" src="https://raw.githubusercontent.com/theopenco/llmgateway/5d0e1e7031cdafb8a4e06bf6f20a289ea5798f78/blog-dark.png" /> </details> ## Reviewer notes — read before approving Three things a reviewer would otherwise have to reverse-engineer: 1. **No shipped change explains the week-over-week jump.** [theopenco#3374](theopenco#3374) (`perf(gateway): kill cold-path TTFT tail` — prewarm pinger, `dnsConfig`/ndots, hedging, CPU limits) is **closed and unmerged**; `UPSTREAM_PREWARM_ORIGINS` is not on `main`. The only merged latency work, [theopenco#3225](theopenco#3225), landed Jul 25 — before _both_ runs, including the Jul 31 one where we placed **last**. The post therefore describes theopenco#3225 as the engineering behind our upstream path and makes **no causal claim** about the Jul→Aug delta. Please keep it that way if you edit the copy. 2. **This is one run, and the field is tight.** Our +1.06 over the baseline is 62% a single number (warm TTFT p95 ≈ 1–2 samples out of 20). The post has a dedicated "Read the Caveats Before You Cite This" section and links the live benchmark rather than embedding a screenshot, so readers see current numbers. 3. **The unflattering history is public in the same repo.** `results/ai-gateway/2026-07-31.json` shows us last with a 3368 ms cold p95. The post says so directly rather than inviting a gotcha. Two claims are deliberately **left out**: that we're faster than calling Anthropic directly (the FAQ states the opposite outright — our warm TTFT _median_, 629 ms, is slower than the baseline's 615 ms), and our field-leading tokens/sec (92.7), since `outputTokensPerSec` is regexed from the SSE buffer on a per-wire-format path and we're compared on the `openai` path against the baseline's `anthropic` path. ## Reproduction steps were verified against the pinned tree The "Run It Yourself" block is pinned to `7548e7584940`, the commit holding the August 7 results. That commit postdates computesdk's restructure into a pnpm workspace, so the instructions use `pnpm install` / `pnpm bench:ai-gateway` and reference `benchmarks/ai-gateway/scoring.ts` (there is no `src/` at that commit). `--iterations 20` is correct, not 40: `ai-gateway.bench.ts` sets `ITERATIONS_COLD` and `ITERATIONS_WARM` _each_ from the flag, so 20 yields the cited 20 cold + 20 warm; the `config.iterations: 40` in the results JSON is their sum. ## Verification - `pnpm format` — clean - `turbo run build --filter=ui` — passes (content-collections validates the frontmatter schema) - `/blog/ai-gateway-benchmark` returns 200 and `/changelog/ai-gateway-benchmark` returns 404 against the production build - Benchmark figures read from `results/ai-gateway/latest.json` at `computesdk/benchmarks@7548e7584940` (`generatedAt 2026-08-07T13:18:32Z`), not transcribed from a screenshot Content-only; no code paths touched. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Luca Steeb <contact@luca-steeb.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.
No description provided.