v0.5.2

Fixed

• up only says "ready" when the stack is ready. On first start Open WebUI downloads its support models, which took longer than the old 45 second wait, so localhost:3000 refused connections right after the "ready" banner. up and update now wait up to 5 minutes by default (change it with --timeout), the progress line names the services still starting, and Open WebUI is checked on its /health endpoint. If a service is still not ready, up says so and shows how to follow it with pondok-ai status and pondok-ai logs. (#37)

Added

• up starts native Ollama on macOS for you. On stacks set up with init --native-ollama, when Ollama is not running, up opens the Ollama app when it is installed, and otherwise runs ollama serve in the background with its log in .pondok-ai/ollama.log. You only get the manual hint when both fail. (#37)
• up tells you which model to install first. When Ollama has no models yet, up prints the recommended model for your RAM (pondok-ai model install ...), since Open WebUI has nothing to chat with until one is installed. (#37)

Install

npx pondok-ai@0.5.2 init

Already have a stack? npx pondok-ai@0.5.2 up picks up the new readiness check, no config change needed.

API Gateway image: docker pull ghcr.io/rasidi3112/pondok-ai-gateway:0.5.2

v0.5.1

Fixed

• Every model can chat in Open WebUI, including Gemma 3. gemma3:4b and gemma3:12b, which pondok-ai model recommend --lang id suggests, failed in Open WebUI with "does not support tools". Open WebUI 0.11 sends its built-in tools with every chat, and Ollama rejects tools for models without tool support. New stacks now start Open WebUI with legacy function calling, which works with every model. (#34)

  Already running a stack? Open WebUI keeps this setting in its database, so change it once as the admin: Admin Panel > Settings > Models > Model Defaults > Configure > Model Parameters > Configure, click Function Calling until it shows Legacy, then Save. See Troubleshooting.

Install

npx pondok-ai@0.5.1 init

API Gateway image: docker pull ghcr.io/rasidi3112/pondok-ai-gateway:0.5.1

v0.5.0

Added

• Classroom mode: pondok-ai share. One machine becomes a private ChatGPT for a whole room. share publishes Open WebUI on your Wi-Fi or LAN and prints its address with a QR code, so a classroom, office or pesantren can chat from their own phones and laptops, each person with their own account. It is safe by default:

  • It refuses to run until the Open WebUI admin account exists, because the first account to sign up becomes the admin.
  • Login is always on, and new accounts wait as pending until the admin approves them. Approved users can chat with the models installed on the machine.
  • Only Open WebUI is shared. The API gateway, Ollama, Qdrant and n8n stay on 127.0.0.1.

  Use --light on light terminal themes so the QR code scans, and pondok-ai share --stop to turn sharing off. (#31)

• Models for Bahasa Indonesia. pondok-ai model recommend --lang id lists only models that answer well in Indonesian and picks one that fits your RAM: Qwen 2.5 (1.5B, 3B, 7B), Gemma 3 (4B, 12B), Aya Expanse (8B), and BGE-M3 embeddings for RAG over Indonesian documents. (#31)

Fixed

• The memory check before a download no longer misjudges larger models. qwen2.5:32b used to be checked as qwen2.5:1.5b and pass on a 16 GB machine. Tags missing from the list are now estimated from their parameter count and quantization. (#31)
• Recreating the Open WebUI container (update, share, config edit) no longer logs everyone out. The session key now lives in the Open WebUI data volume. (#31)

Install

npx pondok-ai@0.5.0 init

Share it with your room:

npx pondok-ai up       # then create the admin account at http://localhost:3000
                       # and turn on New Sign Ups in Admin Panel > Settings > General
npx pondok-ai share

API Gateway image: docker pull ghcr.io/rasidi3112/pondok-ai-gateway:0.5.0

v0.4.0

Added

• Native Ollama on macOS for the Apple Silicon GPU (Metal). Docker on macOS cannot reach the GPU, so pondok-ai init --native-ollama (or answering yes when an Apple Silicon GPU is detected) skips the Ollama container and points Open WebUI and the API Gateway at the Ollama installed on your Mac. model commands, health checks and up use it in this mode. Linux and Windows behavior is unchanged. (#12)
• 1080p demo video, demo GIF and HD screenshots in the README, with the scripts used to record them in docs/demo/. Everything shown is a real CPU-only run. (#13, #14)
• Polish README translation. Thanks @MeydhiXploit! (#16)
• API Gateway image on GitHub Container Registry for amd64 and arm64: docker pull ghcr.io/rasidi3112/pondok-ai-gateway:0.4.0. The CLI still builds the gateway locally. (#15)
• Bug report and feature request forms, Dependabot updates and CodeQL scanning. (#15)

Changed

• The NVIDIA GPU prompt no longer appears on macOS. (#12)
• README restructured: centered header with badges, screenshots, a feature table and a shorter quickstart. (#14)
• All 16 translated READMEs synced with the English README, with the same language selector in every file. (#30)
• Security reports now go through GitHub Private Vulnerability Reporting. (#15)

New contributors

• @MeydhiXploit made their first contribution in #16

Install

npx pondok-ai@0.4.0 init

On an Apple Silicon Mac with Ollama installed: npx pondok-ai@0.4.0 init --native-ollama

v0.3.1

Security

• Updated the gateway's web framework from Fastify 4.29.1 to 5.12.5 (major upgrade) and @fastify/cors to 10.1.0 to address published advisories (request validation and authentication bypass issues, and denial of service). Gateway behavior is unchanged: Bearer auth, rate limiting, CORS and error formats stay the same.

Added

• CI now also checks streaming responses (stream: true returns data: events ending with [DONE]).

Install

npx pondok-ai@0.3.1 init

Release v0.3.0

BREAKING

• Renamed from local-ai-stack to pondok-ai: npm package pondok-ai, CLI command pondok-ai, config directory .pondok-ai, container names pondok-ai-*. Existing 0.2.x stacks are not migrated automatically. The old package local-ai-stack is deprecated.

Install

npx pondok-ai init

Notes

• "Pondok" is Indonesian for a small lodge or shelter.
• End-to-end CI runs on amd64 and arm64. Known limitations are listed in the README.

v0.2.4

Fixed

• up no longer fails to start the stack: healthchecks for Ollama, Qdrant and the gateway no longer use curl, which those images do not include.
• Qdrant's healthcheck only verifies that port 6333 is open.
• The tag-triggered release workflow installs the gateway dependencies before running tests, and no longer publishes (publishing is manual).

Added

• CI end-to-end smoke test on amd64 and arm64: starts the stack and checks 401 without a key, rate limiting (429 + Retry-After) and /health.

Note

• 0.2.0 through 0.2.3 are deprecated: up could not start the stack in those versions.

Install

npx local-ai-stack@0.2.4 init

v0.2.2

Fixed

• up no longer fails to start the stack: healthchecks for Ollama, Qdrant and the gateway no longer use curl, which those images do not include.
• Qdrant's healthcheck only verifies that port 6333 is open.
• The tag-triggered release workflow installs the gateway dependencies before running tests.

Added

• CI end-to-end smoke test on amd64 and arm64: starts the stack and checks 401 without a key, rate limiting (429 + Retry-After) and /health.

Note

• 0.2.0 and 0.2.1 are deprecated because up could not start the stack.

v0.2.1

Fixed

• up no longer fails to start the stack: healthchecks for Ollama, Qdrant and the gateway no longer use curl, which those images do not include.
• Qdrant's healthcheck only verifies that port 6333 is open.

Added

• CI end-to-end smoke test on amd64 and arm64: starts the stack and checks 401 without a key, rate limiting (429 + Retry-After) and /health.

Note

• 0.2.0 is deprecated because up could not start the stack.

v0.2.0

Breaking changes

• Minimum Node.js is now 22 (engines.node >= 22).
• API Gateway /v1/* requests require Authorization: Bearer <API_SECRET_KEY>. Update any client that sends a placeholder key.

Added

• Configurable in-memory rate limiting for /v1/* (RATE_LIMIT_MAX_REQUESTS, default 120; RATE_LIMIT_WINDOW_MS, default 60000). Over-limit requests receive 429 with Retry-After.

Changed

• Docker images pinned to explicit versions. update refreshes pins; update --upgrade-images migrates legacy default tags without overwriting custom values.
• Installer URLs and package metadata now point to this repository.

Fixed

• Backup/restore docs corrected: backups are integrity-checked with SHA-256, are not encrypted, and do not provide cryptographic authenticity guarantees.

Install

npx local-ai-stack@0.2.0 init