Blocker-aware decision layer for AI agents

Your AI sounds confident.
It missed the blockers.
PocketLantern caught them.

PocketLantern attaches source-linked blocker facts to AI technical decisions.

View on GitHub Read the Docs

Open-source local toolkit now. Hosted Pro coming soon.

Caught recently: Next.js 16 custom webpack breakage · OpenAI preview model removal · Cognito hash export limits

MCP client — PocketLantern search_cards
you › |
⟶ search_cards(query: "Clerk Auth0 Cognito", include_blockers: true)
✓ auth/clerk-vs-auth0-vs-cognito-2026 + 4 cards
Lock-in: vendor/cognito (high) — password hashes permanently locked
Export: vendor/auth0 → csv (partial) — support ticket required
Lock-in: vendor/auth0 — Rules/Hooks EOL November 2026
───
AI › Clerk: hashes exportable as bcrypt — lowest lock-in.
     Cognito: hashes can never be exported. Highest lock-in.
     Auth0: export requires support ticket, not on Free.
     ···

Simulated from actual search_cards output. Source-linked facts, reviewed March 2026.

Ask your AI about choosing auth, upgrading frameworks, or using new APIs. It sounds fine.
PocketLantern surfaces the blockers it may miss.

"Which auth provider — Clerk, Auth0, or Cognito?"

Without PocketLantern

"It depends on your team size. Clerk for DX, Auth0 for enterprise, Cognito for AWS cost savings."

  • No migration cost analysis
  • No password export limitation
  • No vendor EOL timeline
Result: Plausible, but missing irreversible lock-in
With PocketLantern

"Factor in migration escape cost before choosing:"

  • Cognito password hashes permanently non-exportable
  • Auth0 export requires support ticket — Free tier excluded
  • Auth0 Rules/Hooks EOL 2026-11-18
Result: Same question, plus irreversible lock-in risks surfaced
Sources
AWS Cognito Developer Guide — User Pool Export reviewed 2026-03 Auth0 Docs — User Import/Export reviewed 2026-03 Auth0 Docs — Migrate from Rules to Actions reviewed 2026-03 Clerk Docs — Export Users (API) reviewed 2026-03

"Should I upgrade to Next.js 16?"

Without PocketLantern

"Turbopack is now default. Use the codemod and upgrade."

  • No webpack build failure warning
  • No sync API removal detail
  • No middleware deprecation
Result: Sounds simple — but builds can break
With PocketLantern

"Upgrade, but plan for 4 breaking changes:"

  • Sync API access removed — every call must be awaited
  • Custom webpack config breaks next build
  • next lint removed — switch to Biome/ESLint
  • Node.js 20.9.0+, TypeScript 5.1.0+, React 19.0+
Result: Same advice, plus 4 blockers and version floors
Sources
Next.js Blog — Next.js 16 reviewed 2026-03 Next.js Docs — Upgrading: Async Request APIs reviewed 2026-03 Vercel Changelog — next lint removal reviewed 2026-03

"Can I use the OpenAI Realtime API for voice agents?"

Without PocketLantern

"GA is available. Use gpt-4o-realtime-preview for production."

  • No model sunset date
  • No event schema incompatibility
  • No auth endpoint change
Result: Suggests a model being removed in 7 weeks
With PocketLantern

"Use GA, but preview models are removed 2026-05-07:"

  • gpt-4o-realtime-preview removed May 7
  • Beta interface (realtime=v1 header) removed same date
  • GA event schema completely different — 4 renames
Result: Hard deadline surfaced — start on GA or plan migration
Sources
OpenAI Docs — Realtime API (GA) reviewed 2026-03 OpenAI Blog — Realtime model deprecation schedule reviewed 2026-03 OpenAI Docs — Beta to GA migration guide reviewed 2026-03

Same question. Same AI. Different facts.
One search. Blockers surfaced.

How It Works

One question. Blockers surfaced. That's the experience.

1

Search

search_cards("clerk vs auth0 authentication")

matched cards + blocker warnings from sidecar blocker edges

2

Blockers

⚠️ Lock-in: Cognito hashes non-exportable · EOL: Auth0 Rules Nov 2026

pipeline-checked constraints that training data often misses

3

Decide

Agent answers with the full picture — with blockers and version floors included

What We Catch

Not opinions. Not tutorials. Not “best practices.”
Breaking changes, EOL dates, lock-in risks, and other blocker facts AI agents often miss.

!

Breaking changes

Next.js 16 webpack build failure, Prisma v7 ESM-only, OpenAI event schema rename

§

License shifts

Redis 8 AGPL copyleft, Terraform BSL 1.1, Valkey as BSD alternative

$

Lock-in & export

Cognito password hashes permanently locked, Supabase pg_dump fully portable

#

EOL dates

OpenAI Realtime Beta May 2026, Node.js 20 April 2026, Auth0 Rules Nov 2026

Curated blocker-dense decision cards. Try these first:

auth/clerk-vs-auth0-vs-cognito-2026

Cognito password hash permanent lock-in, Auth0 export restrictions, Rules/Hooks EOL timeline.

Clerk · Auth0 · Cognito
frontend/nextjs-16-upgrade-turbopack-default-and-async-request-apis

4 simultaneous breaking changes, webpack build failure, middleware deprecation, version floors.

Next.js 16 · Turbopack · React 19
ai/openai-realtime-api-beta-to-ga-migration

Beta + preview model removed 2026-05-07, GA event schema incompatible, auth endpoint moved.

OpenAI Realtime · WebRTC · Voice AI
Browse all cards →

What's Inside

Measured from the actual bundle shipping with this repo.

1,960 Pipeline-checked blocker facts across real technical decisions
100% of bundled cards include source-linked blocker data
14 Blocker types — EOL, lock-in, breaking changes, version floors
731 Vendor lock-in warnings — export blocks, SDK coupling, portability gaps
128 EOL dates tracked — model shutdowns, runtime deprecations, API sunsets
0 Third-party blog sources — official vendor docs and changelogs only

Get Started

Clone the open-source toolkit and run it locally with Claude Code, Cursor, or any MCP client.

$ git clone https://github.com/pocketlantern/toolkit
$ cd toolkit && pnpm install && pnpm build
$ npx pocketlantern init   # registers MCP server
# Then ask: "Which auth provider — Clerk, Auth0, or Cognito?"

100+ curated blocker-aware cards across 25 categories. Sidecar blocker edges included where bundled.

GitHub Docs Newsletter — coming soon