LlamaIndex + MinTok Integration
RAG for codebases without drowning in unindexed boilerplate.
How LlamaIndex Works with MinTok Edge
When integrating LlamaIndex with MinTok, your development workflow benefits from a two-tier edge compilation architecture. Requests originating from LlamaIndex route through MinTok's Cloudflare Edge proxy (`https://mintok.adstim.net/v1`).
Before payload transmission to the underlying reasoning models, the edge compiler validates incoming JSON schemas, isolates tool call histories, and applies Tree-sitter abstract syntax tree compaction to file reads. This ensures that LlamaIndex maintains a lightweight context footprint even during complex, 50-turn debugging sessions across enterprise repositories.
Step-by-Step Configuration
Configuration Payload
from llama_index.llms.openai import OpenAI
from llama_index.core import Settings
Settings.llm = OpenAI(
model="mintok-1-max",
api_base="https://mintok.adstim.net/v1",
api_key="mtk_live_xxxxxxxxxxxxxxxx"
)
# Custom Node Postprocessor for AST Slicing
from llama_index.core.postprocessor import BaseNodePostprocessor
class MinTokASTPostprocessor(BaseNodePostprocessor):
def _postprocess_nodes(self, nodes, query_bundle):
# Strips comments and dead branches before synthesis
return [n for n in nodes if len(n.get_content()) > 40]
Dual-Model Strategy: Maximize Speed & Slash Unit Cost
To maximize cost efficiency and speed in LlamaIndex, we recommend a dual-model configuration:
1. Fast / Agile Tier (`mintok-1-flash`): Set as your inline autocomplete, git commit message, and test-runner summarization model. Operates at <150ms latency and $0.05/M prompt tokens.
2. Frontier / Architecture Tier (`mintok-1-max`): Set as your primary agent reasoning and multi-file refactoring model. Captures 1658 Coding Arena Elo at $0.42/M prompt tokens (6.2x cheaper than Claude Sonnet 3.7).
Common Edge Cases & Resolution
Resolving Context Window Truncation
Ensure your request payload does not manually duplicate previously sent file contents. MinTok automatically handles state deduplication across turns.
Custom Proxy & Corporate Firewalls
If your corporate network intercepts SSL certificates, ensure the MinTok base URL is whitelisted: `https://mintok.adstim.net/v1`. MinTok runs on Cloudflare's global anycast network with TLS 1.3.
Streaming Response Buffer Stalls
If completions appear buffered in your terminal or IDE, ensure HTTP/2 or SSE streaming is enabled in your configuration.
Verification & Connection Test
Once configured, verify connectivity with: curl -H "Authorization: Bearer mtk_live_..." https://mintok.adstim.net/v1/models
Everything You Need to Know About LlamaIndex + MinTok
How long does it take to connect MinTok to LlamaIndex?
Setup typically takes under 2 minutes. You only need to set the OpenAI Base URL to `https://mintok.adstim.net/v1` and add your `mtk_live_...` API key in LlamaIndex's settings.
Are there rate limits on concurrent agent loops?
Standard plans include up to 100 requests per minute with automatic burst scaling across Cloudflare Edge PoPs.
Can I use MinTok with private corporate repositories?
Yes. MinTok enforces zero code retention. Code is processed inside secure V8 WebAssembly sandboxes and never written to disk or used for training.
What credit balance do I need to get started?
Every new account receives $10 in free testing credits, which is sufficient for over 10,000 tasks on MinTok-1-Pro.