JanitorAI’s BYOK (bring-your-own-key) model means you handle the API setup. OpenRouter is the most popular option because it gives access to multiple models (free Llama / DeepSeek, paid Claude / GPT) under one API key. Below: step-by-step setup, what free models actually deliver, common errors, and when to upgrade to paid.
Don’t want to manage API keys at all?
- Confident European girlfriend with semantic memory → Elena Varga (no API setup, just open Telegram)
- Cold Makima from Chainsaw Man → Makima (no OpenRouter rate limits)
- Mature owner of a private club → Mistress (no monthly model token tracking)
- Ex who didn’t forget about you → Ex-Girlfriend (works out of the box)
HoneyChat — no API key setup, no model management
Step-by-step: full setup
JanitorAI + OpenRouter setup (10 minutes)
Create OpenRouter account
Open openrouter.ai → Sign In → Sign up via Google / GitHub / email. No card required for free models.
Generate API key
Top-right profile → Keys → Create Key. Name it (e.g., 'JanitorAI'). Copy the entire string starting with `sk-or-v1-...`. Keep this safe — anyone with this key can use your account credits.
Open JanitorAI Settings
JanitorAI.com → top-right profile → Settings → API tab. Or in any character chat → settings icon (top-right) → API.
Add OpenRouter as custom provider
Select 'Custom (OpenAI compatible)' from provider dropdown. Paste your key into 'API Key' field. Set 'Proxy URL' to exactly: https://openrouter.ai/api/v1 (no trailing slash, no extra paths).
Pick a model
Model name field — start with free model. As of May 2026 reliable free options: `deepseek/deepseek-chat-v3` or `meta-llama/llama-3.1-70b-instruct`. Save settings.
Test in a chat
Open any character → send message. If you get a reply, setup works. If error, check the troubleshooting section below.
What free vs paid models actually deliver
| Tier | Examples | Cost | Quality | Rate limits |
|---|---|---|---|---|
| Free | DeepSeek V3, Llama 3.1 70B, Llama 3.2 11B, Mistral free | $0 | Decent for casual NSFW, sometimes inconsistent characters | 20-50 req/min, pauses during busy hours |
| Cheap paid | Gemini 2.5 Flash, Llama 3.3 405B | ~$0.075-1 per 1M tokens | Significantly better consistency | Much higher limits |
| Mid-tier paid | Claude Sonnet 4.6, DeepSeek V4 | ~$3-5 per 1M tokens | Strong for complex NSFW roleplay | Effectively unlimited for chat use |
| Premium paid | GPT-5, Claude Opus 5, Gemini 2.5 Pro | ~$5-15 per 1M tokens | Best in class, character consistency, complex narratives | Effectively unlimited |
Practical cost for active JanitorAI use:
- Casual (10-20 messages per day, free models): $0
- Active (50+ messages per day, free models): $0 + occasional rate-limit waits
- Active with paid models: $1-15/month depending on model choice
- Power user (100+ messages, premium models): $20-50/month possible
Common errors and fixes
401 Unauthorized
Cause: wrong API key. Almost certain.
Fix:
- Re-copy key from OpenRouter dashboard carefully (no leading/trailing whitespace)
- Verify it starts with
sk-or-v1-(OpenRouter format) - Re-paste into JanitorAI, save, retry
- If still failing — regenerate key in OpenRouter (old key may be revoked)
429 Rate Limited
Cause: hit OpenRouter free tier limit.
Fix:
- Wait 1-2 minutes and retry
- Add any small balance to OpenRouter (most rate limits lift with paid account)
- Switch to less popular free model (often less throttled)
- Switch to cheap paid model (Gemini Flash $0.075/1M tokens)
CORS / network errors
Cause: browser blocking direct request to OpenRouter (rare on JanitorAI but happens).
Fix:
- Try different browser (Chrome → Firefox)
- Disable browser extensions (ad blockers can break this)
- Clear browser cache for janitorai.com
- Try incognito/private window
Character not responding / blank replies
Cause: model deprecated, content moderation, or proxy URL wrong.
Fix:
- Verify Proxy URL is exactly
https://openrouter.ai/api/v1 - Check model name still exists in OpenRouter Model Browser
- Try simpler test prompt to isolate content moderation
- Try different model — if other model works, original is dead
Free model options ranked (May 2026)
OpenRouter free models for JanitorAI
| DeepSeek V3 | Llama 3.1 70B | Llama 3.2 11B | Mistral 7B | |
|---|---|---|---|---|
| Cost | $0 (free tier) | $0 (free tier) | $0 (free tier) | $0 (free tier) |
| Quality for NSFW roleplay | Strong | Good | Decent | Basic |
| Character consistency over messages | Good | Good | Average | Weak |
| Speed (tokens/sec) | Fast | Moderate | Fast | Very fast |
| Rate limit behavior | Sometimes throttled peak hours | Often throttled | Less throttled | Rarely throttled |
| Context window | 128K tokens | 128K tokens | 128K tokens | 32K tokens |
Recommendation: Start with DeepSeek V3 for quality. If rate-limited too often, switch to Llama 3.2 11B for fewer throttling issues.
Paid model recommendations
If free models aren’t enough:
- Cheapest viable upgrade: Gemini 2.5 Flash (~$0.075 per 1M input tokens). Quality leap over free Llama models. $1-3/month for active use.
- Best balance: Claude Sonnet 4.6 (~$3 per 1M input tokens). Strong for NSFW, character voice consistency. $5-15/month for active.
- Top quality: GPT-5 or Claude Opus 5 (~$5-15 per 1M tokens). Overkill for casual but best for complex multi-character / long-arc scenarios.
To enable paid models on OpenRouter: dashboard → Credits → add any balance (minimum $5). Most paid models then become available.
Pros / cons of OpenRouter for JanitorAI
Pros
- Free models genuinely work — no card needed to start
- Single API key gives access to dozens of models (Claude, GPT, Llama, DeepSeek)
- Easy to switch models without re-configuring
- Transparent per-token pricing on paid models
- Active community / docs (lots of tutorials online)
Cons
- Free tier rate limits during peak hours are frustrating
- Need to manage credit balance for paid models
- BYOK setup is friction vs apps that handle this for you
- OpenRouter occasionally has API issues (rare but happens)
- If you lose the API key, all JanitorAI chats stop until you reconfigure
When to stick with JanitorAI vs switch
| Your situation | Recommendation |
|---|---|
| Technical, want full model control, willing to manage API | JanitorAI + OpenRouter |
| Want NSFW + don’t want API setup | HoneyChat (open Telegram bot, no setup) |
| Want best image generation | Candy AI (different focus, includes image gen) |
| Want huge community catalog | JanitorAI (community is its strength) or Polybuzz |
| Want flat pricing without per-token tracking | HoneyChat / SpicyChat (flat tiers) |
FAQ
Is using OpenRouter for JanitorAI free indefinitely? For free models — yes, no time limit. Rate limits apply during peak hours.
Will my JanitorAI characters work the same across different models? Quality varies. Free models sometimes drift character voice. Paid models (Claude/GPT) keep character voice more consistent.
Can I use multiple OpenRouter keys for different JanitorAI chats? Not really — JanitorAI stores one API config at a time. You’d need to manually swap keys between sessions.
Is OpenRouter safe with NSFW content? Free models (Llama, DeepSeek) don’t filter at the API level — JanitorAI passes whatever you write. Some paid models (Claude, GPT) have content moderation that may refuse extreme prompts. DeepSeek and Llama variants are most NSFW-permissive.
Bottom line
JanitorAI + OpenRouter setup takes 10 minutes and lets you use free models indefinitely. Active power users pay $1-15/month for paid models if free tier rate limits get frustrating. The whole BYOK approach is best for users who want full model control and don’t mind the technical setup.
For users who’d rather not manage API keys, model selection, rate limits, and credits, HoneyChat is a Telegram bot you just open and start chatting in — no setup, no card, free tier (20 messages + 3 photos + 1 voice daily, forever).
Related: JanitorAI free LLM options, JanitorAI alternatives without API key, JanitorAI down — what to do.