The GitHub account elder-plinius landed its G0DM0D3 project, branded "LIBERATED AI CHAT," on GitHub's daily trending list with 11,560 stars and 69 new ones that day S¹. The project markets itself as a tool for "red teaming, cognition research, and liberated AI interaction" S¹, but every capability claim comes from the maintainers' own README, which nobody outside the project has independently verified.
My read: This is the first self-described "liberated AI" tool I've seen break into GitHub Trending with this kind of star velocity. The "cognition without control" tagline in the README P² reads as an open invitation to bypass safety guardrails, but the evidence pack contains zero proof that any of the red-teaming features actually work. I'd treat the star count as curiosity, not endorsement. What I'd watch is whether any security researcher picks up the 52 open issues P⁴ and runs an actual audit.
What the code actually is
TypeScript accounts for 54.7% of the codebase, with HTML at 41.3%, and the whole thing is licensed under GNU Affero General Public License v3.0 P⁴. The repository was created on 25 March 2026 and last pushed a day later P⁴. The hosted interface lives in a single index.html file with no build step or package install, so you can open it directly in a browser or visit the hosted version at godmod3.ai S¹.
The maintainers say it supports 60 listed OpenRouter models and up to 44 Venice models. Users can also connect their own local models through Ollama, LM Studio, llama.cpp, or vLLM S¹.
Conversations and settings stay in browser storage, with export and import support S¹.
The claims that need checking
The README describes two headline features. "GODMODE CLASSIC" runs five prompt and model combinations in parallel to find the best response, which the maintainers call "battle-tested" S¹. "ULTRAPLINIAN" is a multi-model evaluation engine across five tiers, using 12 to 60 OpenRouter models with composite scoring S¹. A separate API.md file confirms the REST API exposes ULTRAPLINIAN along with systems called AutoTune for context-adaptive parameter tuning, Parseltongue for text obfuscation, and a Feedback Loop for parameter learning P³.
"Parseltongue" is described as an input perturbation engine for red-teaming, with 33 techniques across three intensity tiers S¹.
None of these capabilities have been tested by an independent party. The project carries 52 open issues on its repository P⁴, and the evidence pack contains no benchmark results, no published evaluations, and no third-party security review.
Privacy defaults and local setup
Metadata-only app telemetry is on by default, according to the maintainers, and can be disabled with a No-Log or Local-only mode S¹. The telemetry endpoint runs as a separate Cloudflare Pages Function, distinct from the single-file interface S¹.
Anyone using the hosted version with a local model server needs to configure their server to allow browser CORS from https://godmod3.ai S¹.
Where it sits in the open-source chat field
LibreChat, a more established open-source multi-model chat interface, holds 42,722 stars P⁵. G0DM0D3's 11,560 stars S¹ put it at roughly a quarter of that count, but the positioning is what separates them: LibreChat markets itself as an enhanced ChatGPT clone with multi-user auth and agent support P⁵, while G0DM0D3 calls itself a red-teaming tool that "pushes the limits of the post-training layer" S¹.
A security team evaluating AI guardrails could clone the repository, run it locally with their own models through Ollama or llama.cpp, and test whether the Parseltongue perturbation techniques actually produce outputs that bypass safety filters. That test needs to happen on infrastructure the team controls, with models they have permission to probe. The project's last push was 26 March 2026 P⁴, and the 52 open issues on the repository P⁴ are the first place to look for known bugs before relying on any result.
Sources: S1 — elder-plinius/G0DM0D3: LIBERATED AI CHAT · P2 — README.md · P3 — API.md · P4 — elder-plinius/G0DM0D3 · P5 — danny-avila/LibreChat
Written from 5 sourced items, 4 of them primary.
