Free, open-source tools for developers who ship solo. Tool-agnostic. Local-first. No accounts. MIT licensed.

Local spend monitor for Claude Code. Watches your session logs, tracks cost per project and git branch, warns you before you hit a cap, and kills the process when you blow past it.
budgetclaw alerts setup/v1/models + daily rate cross-check against the community-maintained LiteLLM pricing data -- new models AND rate changes surfaced within 24 hours$ budgetclaw status PROJECT BRANCH TODAY WEEK myapp main $4.85 $28.30 myapp feature/auth $1.20 $3.20 side-proj main $0.00 $0.50 TOTAL $6.05 $32.00 $ budgetclaw limit list # PROJECT PERIOD CAP ACTION 1 * daily $10.00 warn 2 myapp daily $5.00 kill
Your personal AI cost bodyguard. Protects you from expensive surprises and notifies you according to your personal budgets. Connect your AI accounts and see all limits and costs in one convenient dashboard.
AI Cost DashboardAnthropic
Claude Opus 4, Sonnet 4
$412.80
this month
ElevenLabs
3,205 / 10,000 characters
32%
of quota
More providers coming
OpenAI, Stability, ...
Validate Claude Code plugin manifests before submission. Catches reserved marketplace names, duplicate hooks declarations, path traversal, and the schema errors the official validator reports opaquely. Go binary, GitHub Action, zero network calls.
判子 · the personal seal stamped on official Japanese documents.
Framework-specific plugins for Cursor. Verified rules, migration skills, and anti-pattern detection -- so the LLM generates code that actually works with the latest framework versions.
Prevents v3 hallucinations. 5 rule files, migration skill, validation skill, and a review subagent.
/tw-migrate -- v3 to v4 project migration/tw-validate -- version mixing detectorDjango 5.x + DRF patterns. N+1 query prevention, model scaffolding, settings management, and 14 anti-pattern rules.
/django-model -- full vertical slice scaffold/django-validate -- project anti-pattern auditSolid Stack, Hotwire, params.expect, built-in auth, Kamal 2. Stops Cursor from generating Rails 7 patterns.
/rails-auth -- built-in auth, no Devise/rails-validate -- Rails 7 pattern detectorPydantic v2 migration, async event loop safety, SQLAlchemy patterns, and 13 anti-pattern rules for FastAPI.
/fastapi-endpoint -- async CRUD scaffold/fastapi-validate -- v1 + async auditCache Components, proxy.ts (renamed from middleware), async-only request APIs, the Data Access Layer pattern, Server Actions with Zod + ownership re-check, three-layer auth. React 19.2.
/nextjs-dal -- scaffold the Data Access Layer/nextjs-migrate-v15-to-v16 -- codemod sequencelog/slog, errors.Is/As + %w wrapping, context-first APIs, http.ServeMux method routing (1.22+), range-over-func iterators (1.23).
/go-modernize -- older Go to 1.24 migration/go-error-wrap -- %v to %w + errors.Is conversionjakarta.* imports, constructor injection, SecurityFilterChain, RestClient, records as DTOs, virtual threads, Testcontainers @ServiceConnection.
/spring-boot-migrate-to-3 -- 2.x to 3.x in 10 steps/spring-boot-testcontainers -- @ServiceConnection wiringTop-level Program.cs, Minimal APIs with TypedResults, IHttpClientFactory, EF Core AsNoTracking + projection, HybridCache, built-in OpenAPI.
/aspnet-migrate-to-9 -- older .NET to .NET 9/aspnet-testing-setup -- WebApplicationFactory + TestcontainersKotlin 2.x with the kotlin-compose plugin, StateFlow + collectAsStateWithLifecycle, Hilt, type-safe Navigation, Material 3, Version Catalogs.
/compose-migrate-views-to-compose -- per-screen migration/compose-modernize-build -- libs.versions.toml + KSP + BOMfor_each over count, remote backend + locking, moved/import/removed blocks, ephemeral resources, OpenTofu state encryption, OIDC for CI.
/terraform-refactor-with-moved -- safe rename + restructure/terraform-migrate-secrets -- HCL to Secrets ManagerSchema-first design, sql template tag for parameter binding, drizzle-zod pairing, Postgres RLS, Testcontainers transaction-rollback tests. Covers 0.45.x stable + 1.0-rc next.
/drizzle-migrate-to-v1 -- 0.x to 1.0-rc with codemods/drizzle-rls -- Postgres RLS scaffold + isolation testsSemantic locators (getByRole over CSS), web-first assertions, test.extend fixtures, setup-project + storageState auth, sharded CI with merge-reports.
/playwright-setup-auth -- per-worker, multi-role, JWT bypass/playwright-visual-regression -- macOS-vs-Linux baseline trapLaravel 13.x on PHP 8.3+. The L11 skeleton (bootstrap/app.php, deleted Kernel.php), L12/L13 deltas (Carbon 3, casts() method, #[Scope], Cache::flexible, automatic eager loading, UUIDv7), Livewire 4 SFC without Volt, Inertia 3.1 deferred props, Pest 4 browser testing.
/laravel-migrate-v10-to-v13 -- Kernel.php to bootstrap, $casts to casts()Vue 3.5.34 + Nuxt 4.4.5. Leads with the Vue 3.5 trio LLMs miss (useTemplateRef, useId, reactive props destructure) and the Nuxt 4 app/ srcDir layout. Pinia 3 setup stores, Vue Router 5, useAsyncData singleton-by-key + shallowRef defaults, server/api with method-suffixed handlers, VueUse 14, Vitest 4.
/vue-nuxt-migrate-to-3-5-and-4 -- pages/components to app/, store/ to stores/, Vuex to Pinia 3Svelte 5.55.7 + SvelteKit 2.60.1. Leads with the Runes shape LLMs still miss ($state, $derived, $effect, $props), the Snippet replacement for slots, callback-prop events, and the SvelteKit 2 deltas (no-throw error/redirect, mandatory cookies path, $app/state). Vite 8, TypeScript 6, Vitest 4, Playwright 1.60, bits-ui, sveltekit-superforms + formsnap + zod.
/sveltekit-migrate-to-runes -- Svelte 4 to Svelte 5 + SvelteKit 1 to 2 in stagesAstro 6.3.3. Targets the post-October-2024 surface area LLMs lag on: Server Islands (server:defer + fallback), the Content Layer API (loader: glob/file, render(entry), getEntry), Actions with zod via astro:schema, Sessions, astro:env, and the Tailwind 4 + @tailwindcss/vite move. Vite 7, TypeScript 6, Zod 4, Node 22.12 floor.
/astro-migrate-to-content-layer -- src/content/config.ts to src/content.config.ts, type:'content' to loader:glob, entry.render() to render(entry), entry.slug to entry.id in stagesReact Router 7.15.1 Framework Mode (formerly Remix, merged Nov 2024). Targets the surface LLMs trained pre-Nov-2024 still get wrong: per-route Route namespace types from ./+types/, plain-object loader returns, data() for status, ServerRouter/HydratedRouter, clientLoader + HydrateFallback pairing, react-router.config.ts, @react-router/dev/vite. React 19.2, Vite 8, TypeScript 6, Zod 4, Node 20.19 floor.
/rr7-migrate-from-remix -- @remix-run/* to react-router and @react-router/* in stages, with the codemod pointerExpo SDK 55 (May 2026) with the legacy architecture fully removed. Targets the SDK 51-54 idioms LLMs still emit: Reanimated 4 worklets (with the react-native-worklets/plugin babel move), FlashList 2.x without estimatedItemSize, CameraView, expo-video / expo-audio split, expo-sqlite/kv-store as the AsyncStorage drop-in, Stack.Protected for auth, file-based +api routes. React 19.2, RN 0.83.6, Node 20.19 multi-LTS floor.
/expo-migrate-from-rn-cli -- bare RN or older Expo SDK to SDK 55 stage by stage; Reanimated 3 to 4 + worklets plugin swap; FlashList 1.x to 2.x; classic FileSystem to new File/Directory APIHono v4 (TypeScript edge web framework). Targets the v3-era APIs LLMs still emit because v4.0.0 shipped Feb 2024 and stripped a long tail: c.jsonT, c.stream as Context method, c.env() function form, c.req.cookie, app.showRoutes, app.handleEvent, hono/middleware barrel, app.head. Pin floors enforced: >= 4.9.7 (CVE-2025-59139 bodyLimit bypass), >= 4.12.18 if rendering JSX SSR. Runs unchanged on Cloudflare Workers, Bun, Deno, Node 20+, Vercel Edge, AWS Lambda.
/hono-migrate-to-v4 -- stage-by-stage v3 to v4 migration with the validator-throws change; /hono-rpc-setup -- hc<AppType> with the load-bearing import-type rule (the #1 RPC bundle-bloat pitfall)One email per launch. No spam. Unsubscribe anytime.
A small collection of free, open-source developer tools for people shipping solo. Every tool is MIT licensed, runs locally, and works with the AI coding stack (Claude Code, Cursor). No accounts, no dashboards, no cloud lock-in.
Independent developers who ship without a team. People leaning hard on AI coding tools and wanting to keep control of cost, data, and process. Terminal-native, skeptical of vendor lock-in, happy to read source.
A ronin was a samurai without a lord. Shorthand here for working independently by choice, without a company behind you. The tools are built for that audience. The forge is where the work gets hammered out.
Yes. Every repo is MIT licensed. No paid tiers, no subscriptions, no waitlists. If a future tool needs hosted infrastructure that costs us money, it gets labeled clearly and stays opt-in. Nothing shipped today requires that.
No telemetry, no crash reports, no usage analytics in any CLI. The only network calls are ones you configure yourself, like your own ntfy topic for phone alerts. The source is auditable in an afternoon.
Today, it doesn't. RoninForge exists to build a reputation first. Revenue comes later from a separate paid product built on top of this work, labeled clearly as such. The free tools stay free and MIT.
RoninForge builds free, open-source tools for developers who prefer to own their stack. Everything runs locally. Every repo is MIT licensed. No accounts, no tracking, no vendor lock-in.