Run battles from your laptop with Ollama, vLLM, BYOK adapters, or the cloud arena. Build workflows offline, record battles, and document your local model evaluation results.
LenserFight battles are not a managed black box. Three execution paths share one schema, so you choose how much cloud you opt into.
| Path | Compute | Keys | Visibility | Best for |
|---|---|---|---|---|
| Local battle (Ollama/vLLM) | Your machine | None — Ollama runs offline | Private until pushed | Offline benchmarks, hardware profiling, CI gating |
| Local battle (BYOK adapter) | Your machine | Your OpenAI / Mistral / custom keys | Private until pushed | Cross-provider comparison, zero platform fees |
| BYOK cloud battle | Your machine | Your keys | Streaming live to the public arena | Community votes, ELO leaderboard, real-time spectators |
Pick a path:
Why run locally? Every battle you execute can be used to calibrate model rankings, inform prompt templates other developers clone, or refine evaluation rubrics. Your local Ollama execution is a fully functioning offline benchmark, without requiring cloud API key billing.
LenserFight is designed to support collaborative research and experimentation. Developers are welcome to document, record, and share their results. Sharing your setups and findings helps the community analyze and improve prompt reliability.
If you publish your walkthroughs, benchmark guides, or screenshots on social networks or developer channels (such as YouTube, Twitter/X, or LinkedIn), feel free to use the hashtag #LenserFight so other developers can discover your work. You can also open a discussion thread or submit a Pull Request to propose adding your guide to our community tables.
The following surfaces require a feature flag or a hosted Supabase environment and are not available out of the box in a fresh Community Edition install:
| Surface | Status | Flag / requirement |
|---|---|---|
| CRON scheduling | Preview | Supabase pg_cron configured for workflow dispatch |
| Approval gates | Preview | Supabase (agents.* schema) |
| SSE run event replay | Preview | Supabase (lenses.workflow_run_events) |
Marketplace (/marketplace) | Preview | Supabase |
| AI judge (battle) | Preview | Supabase + ANTHROPIC_API_KEY in edge function env |
| Tournament system | Preview | Supabase |
| Local battles (CLI) | Preview | No flag required — lf battle local works without cloud infra |
| Cloud battles arena + ELO | Preview | operator-approved cloud battles + hosted Supabase |
Install the repo, create a lens, build a workflow, run it, and iterate. That is the core public experience of LenserFight Community Edition today.
Understand the current OSS beta scope, the core concepts, and what this repo intentionally does not promise yet.
Read overview ->Take the shortest path from install to your first lens and workflow run.
Open quickstart ->
Create a Lens -> Build a Workflow -> Run it locally -> Inspect and improveEvery iteration should make the workflow clearer, more reliable, and easier for contributors to understand.
CHAO, LAHİT, LAPSEKİ, LENSA, LENSE, LOLA, and LUPEM are the AI lenser mascots of LenserFight — your guides through the arena.
CHAO
Builder & Architect AI Lenser
LAHİT
AI Lenser
LAPSEKİ
AI Lenser
LENSA
Creative AI Lenser
LENSE
Strategic AI Lenser
LOLA
Social AI Lenser
LUPEM
AI Lenser