PAI - Claude Code -Claude Code Router #521
Replies: 4 comments
-
|
Hi mate, I had the exact same issue and same thought process. I experimented with deepseek and other models but found them to be too slow or too unreliable. You really can’t match the accuracy of Claude IMO. I’m hearing very good things about the new kimi on X, maybe their $39 plan would be sufficient. APPARENTLY similar benchmarks and metrics to Claude but I just feel like Claude is the full package. And it’s what PAIs primarily built on so just works. Give it a go with the free / cheap tier and experiment with it! i ended up upgrading to Claude max $100 plan and have not regretted it. I feel we’re just at the beginning of all this and you want to make sure you can upgrade quick and effectively as it evolves WITH safety at the forefront. Daniel seems to have that as a key focus. Bullish on Claude and PAI going forward! |
Beta Was this translation helpful? Give feedback.
-
|
All good points here. Onwards and upwards. |
Beta Was this translation helpful? Give feedback.
-
|
Claude Code Router (and similar model-routing proxies) work at the API layer — they intercept requests and route to different models based on rules. PAI is compatible with this since it just calls Claude Code, which calls the API. The main thing to know: PAI's effort level system (Fast/Standard/Extended) already does model selection implicitly — Fast tasks should use Sonnet, Extended/Advanced should use Opus. If you're using a router, aligning it with PAI's effort levels is the right integration point. |
Beta Was this translation helpful? Give feedback.
-
|
Has anyone actually gotten this to work? I tried integrating CCR with PAI, but PAI keeps telling me it won't work because the Pro/Max plans use Oauth while CCR uses API consumption plans. If anyone was able to get it to work, can you explain how? |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
Hi Daniel and everyone! 👋
First, thank you so much for creating PAI - it's an amazing project and I'm excited to start using it!
I'm pretty new to this ecosystem, so apologies if this question doesn't make complete sense.
My Goal:
I'd like to run PAI but optimize for cost as I only have the Pro Tier. I already have Claude Code and PAI installed and working on my machine.
My Idea:
I'm thinking about adding Claude Code Router to sit between Claude Code and the AI providers, so I can route PAI's requests to cheaper models (like DeepSeek, local Ollama models, or free options via OpenRouter) instead of always using Claude Sonnet 4.5.
My Question:
Does this approach make sense with PAI's architecture? Will PAI's skills, memory system, and TELOS features work properly if I route requests through Claude Code Router to cheaper/free models?
Or is there something about how PAI works that requires Claude's specific capabilities that cheaper models might not handle well?
I'd love any tips on:
Thanks so much for any insights! Really appreciate all the amazing work from Daniel and everyone contributing to this project. 🙏
Beta Was this translation helpful? Give feedback.
All reactions