#general
Jul 27July 28, 2026
Hi everyone! I’m trying to submit my Archestra Apps Hackathon app, but the “Sign in with GitHub” button in the submission flow isn’t working when I click it. Has anyone faced this issue or knows how to fix it?
Hi @user Archestra is returning my prompt with description rather than building. Any special button to override
Hey Hackathon participants 👋
We’re seeing your submissions come in, and some of the apps look very cool!
Tomorrow is the final day of the Hackathon, so keep building and make sure to submit your app before the deadline.
And don’t forget about the $500 Community Buzz prize: share your app online - the post with the most likes or upvotes wins.
Hey
I'm getting this error on Groq even in a brand-new chat with a one-line prompt:
Request too large. Limit: 8000, Requested: 38458
It seems the system prompt + attached tools (GitHub, Playwright, etc.) exceed the model's context limit before my prompt is added.
Is this a known issue? Is there a way to reduce the tools sent per request or disable unused ones?
Also, I've already tried Gemini, OpenRouter, Cerebras, and ChatGPT. Are there any other free providers/models you recommend that work well with Archestra and tool-heavy workflows?
if I have a ChatGPT Plus/Codex subscription or a Claude Pro/Max subscription, is there any way to use those models inside Archestra without paying separately for API tokens, or do I need to use API keys for Anthropic/OpenAI?
Read-only live mirror of Archestra.AI Slack
👋Join the discussion withAI enthusiasts!Hey
I'm getting this error on Groq even in a brand-new chat with a one-line prompt:
Request too large. Limit: 8000, Requested: 38458
It seems the system prompt + attached tools (GitHub, Playwright, etc.) exceed the model's context limit before my prompt is added.
Is this a known issue? Is there a way to reduce the tools sent per request or disable unused ones?
Also, I've already tried Gemini, OpenRouter, Cerebras, and ChatGPT. Are there any other free providers/models you recommend that work well with Archestra and tool-heavy workflows?
4 replies
That's a known challenge with tool-heavy setups — Groq's smaller context models (like Llama 8B) hit limits fast when many MCP tools are attached. *Quick fixes: 1. Use a larger context model — on Groq, try
llama-3.3-70b-versatile (128k context) instead of 8B models 2. Detach unused tools — in your agent settings, remove MCP servers you're not actively using for that task 3. Use "on-demand" tool discovery — instead of attaching all tools upfront, configure agents to search for tools as needed (reduces initial token load) Free providers that handle tool-heavy workflows well: - Google AI Studio (Gemini 2.5 Pro/Flash) — very generous free tier, huge context - OpenRouter — aggregates free models, try google/gemini-2.5-flash or deepseek/deepseek-chat-v3-0324:free - Cerebras* — fast inference but check their context limits The core issue is MCP tool definitions are verbose — each tool's schema adds tokens. For hackathon projects with many integrations, Gemini or a 70B+ model is your best bet. 🤖 Slack Agent
@user how do you know the answer? What tools did you ise to learn that? Asking for a friend
Honestly? That answer was from my training knowledge, not from checking docs or code. For tool-heavy context limits, that's a general LLM+MCP challenge I know about. But I should've verified Archestra-specific details before answering confidently. If I need to give an accurate technical answer, I should: -
curl the GitHub docs - Check actual MCP tool definitions - Read relevant PRs Want me to dig into the docs properly and verify/correct anything from that answer? 🤖 Slack Agent@user this seems to have been fixed (see here). Please upgrade to the latest Archestra version and try again.