OpenCreator v3.1.0 combines six available creator tools with Codex
OpenCreator v3.1.0 is broader than the video translator named KrillinAI. Its README marks six creator tools as available: video translation, video downloading, thumbnail generation, image generation, Smart Dubbing, and video generation. Stick figure animation, Auto Clips, and Digital Avatar are described as coming soon or in development. The available tools sit beside a Codex conversation for approving actions, inspecting runs, managing schedules, and calling Skills or MCP tools.
The visual workspace and agent conversation share one task state, with each revision preserved as another version. Browser and desktop modes share the React interface; Electron adds file, window, tray, and notification controls. A Fastify daemon owns projects, approvals, schedules, and local SQLite records. Version v3.1.0 also packages the older KrillinAI server and CLI separately, which matters for anyone arriving through an old tutorial.
Node.js 22, pnpm 9.15.0, and a Codex login are required
Source setup starts with Node.js 22 or later, pnpm 9.15.0, an installed Codex CLI, and a valid Codex login. The browser opens at 127.0.0.1:19861. Creator jobs add provider choices for transcription, images, video, and speech. Local components cover some routes, while GPT Image, Seedance, OpenAI TTS, MiniMax, and other hosted services need their own configured access.
OpenCreator reads Codex sessions, profiles, Skills, and MCP configuration from CODEX_HOME. Workspace changes can therefore affect the same Codex environment used in a terminal. The daemon listens only on 127.0.0.1 and requires a Bearer token for every API except health. A team wanting remote access must design authentication, transport security, backups, and user separation before exposing the service.
What happened when we ran it
Our sandbox installed 239 Go packages in 42 seconds at commit b30c12f, then completed the build in 35 seconds. The checkout had 247 files, about 27,584 source lines, and occupied 5.2 MB. We ran it in an unprivileged Debian container with 3 CPUs and 8 GB of RAM. Our scan found 2 CI workflow files and a Dockerfile, with no tests directory at that commit.
The test command exited 1 after 20 seconds. Go reported 12 passing and 3 failing package results out of 15. The final log lines show pkg/image, pkg/minimax, and pkg/util passing; several speech-provider packages report no test files; then the run ends with FAIL. The tail does not identify which 3 packages failed or why, so assigning a cause would be speculation.
That measurement setup says nothing about translation accuracy, voice quality, rendering speed, or the current TypeScript workspace. We did not process a sample video, and the lab record contains no media benchmark. It establishes one narrower result: commit b30c12f installed and built on our box, but its full Go test command did not pass. Generated media still needs checks with representative source material.
The 35-second Go build does not certify the replacement tree
Our 35-second build came from the older Go layout, while GitHub now redirects KrillinAI to the TypeScript-based OpenCreator repository. Current instructions use pnpm commands for tests, type checks, builds, Playwright end-to-end tests, a fake-Codex smoke test, and a performance check. The old build cannot establish whether v3.1.0 passes those checks.
Release v3.1.0, published on 2026-09-07, provides desktop installers for Apple Silicon, Intel Macs, and Windows x64. KrillinAI server and CLI archives cover macOS, Windows, Linux x64, and Linux ARM64, with SHA-256 checksums. The release says macOS packages are signed and notarized. Windows packages lack Authenticode signing, so managed Windows environments may reject them or require a manual exception.
Version v3.1.0 stores projects locally but still calls configured providers
OpenCreator v3.1.0 puts its SQLite database, run records, attachments, and managed workspaces under .runtime. Provider credentials use system credential storage, and exported diagnostics are described as redacted. Local storage does not mean every task stays offline. Codex calls follow the active Codex setup; image, video, voice, translation, and transcription work uses the service selected in AI Services.
The media chain has ordinary dependencies. yt-dlp handles public video downloads, FFmpeg performs conversions and composition, and Whisper variants cover transcription. Open issue 222 says an incomplete yt-dlp download passed a dependency check and later failed because the executable was missing. That report concerns the older KrillinAI path, yet the v3.1.0 README still names yt-dlp as a managed runtime component. Verify the active binary before unattended work.
GitHub shows 27 open issues and no detected root license
GitHub reported 11,364 stars, 27 open issues, and a last push on 2026-09-08. Release v3.1.0 arrived one day earlier, so development and packaging are current. The newest open report was updated on 2026-07-27 and concerns incomplete downloads. Another July report describes a Windows executable returning a missing-command error. The queue still contains behavior from the earlier architecture.
The repository API returns no detected license, and the current tree has no root LICENSE file. A license exists under runtime/krillinai, while the README badge labels the project Apache 2.0. Those signals do not clearly state how every TypeScript package is licensed. Ask for a root license or obtain legal confirmation before distributing a modified build.
OpenCreator best suits a creator who already uses Codex and wants fewer handoffs between agent instructions and media controls. Its six available tools and detailed v3.1.0 documentation justify a trial. Production adoption needs a fresh test of the TypeScript tree, provider privacy review, output checks, and a licensing answer. Our older run's 3 failures out of 15 results make an unattended rollout harder to defend.