Context infrastructure for AI.
We solve the hardest problem in AI development: making sure your AI actually understands your work. Semantic search, automatic context routing, and intelligent memory—across any LLM.
The calibration threshold is updated in three places:
1. update_threshold() in calibration.py:187 is called after every miss — it lowers the similarity floor by the configured step size.
2. AdaptiveRouter.recalibrate() in cloud_router.py:44 reads the new threshold and re-weights candidate scoring on the next query.
3. The floor and ceiling bounds live in config.py:23 as CALIBRATION_MIN and CALIBRATION_MAX — they prevent runaway drift.
You ask a question. Pyckle handles the rest.
Simulated output. Token savings vary by codebase and query.
The Problem
of tokens can be wasted on irrelevant context
more time spent hunting for files without context
memory between sessions
Every AI coding tool uses the same models. The difference is context.
We make context automatic.
Our Approach
Find code by meaning, not keywords. Ask "where do we handle auth?" and get the right files—not string matches.
Every prompt gets the right code snippets automatically. No manual @mentions. No wasted tokens. ~50ms typical latency.
Your code stays on your machine. Free tier runs entirely local—no API keys, no cloud dependency, no trust required.
Architecture
MCP Tools
Everything your AI needs to understand your codebase.
Semantic code search by meaning, not keywords.
Learn more →Index your codebase for instant semantic queries.
Learn more →View indexing statistics and health metrics.
Learn more →Track token usage and optimize context budget.
Learn more →Resume sessions with full context memory.
Learn more →Generate summaries of session activity.
Learn more →Track file edits for context coherence.
Learn more →Explore code dependencies and connections.
Learn more →Analyze blast radius of code changes.
Learn more →Index an Obsidian vault for semantic retrieval.
Learn more →Index a Notion database or page for semantic queries.
Learn more →Query commit history with natural language.
Learn more →Model-agnostic prompt routing for any MCP client.
Learn more →Start an autonomous goal-directed iteration session.
Learn more →Record an iteration result — keep, discard, or crash.
Learn more →Get live progress and metrics for an active loop.
Learn more →List all autoloop sessions for a codebase.
Learn more →Mark an autoloop as complete and archive results.
Learn more →Index GitHub and GitLab issues for semantic search.
Learn more →Store a decision or insight in persistent memory.
Learn more →Retrieve past decisions and context from memory.
Learn more →Map source files to their test coverage.
Learn more →Products
Context infrastructure for every AI workflow.
Everything included. One price.
Semantic context for every AI prompt — invisible, ~50ms, no workflow changes. Works natively with Claude Code and any MCP-compatible editor.
PyckLM-powered code vectors
Production-grade code embeddings via REST API. PyckLM understands code structure — not just text. Drop-in for any app that needs semantic code search.
The AI router that knows your code
Multi-provider AI routing with automatic codebase search. Route queries to the best model — Anthropic, OpenAI, Groq, Mistral, Ollama — with your codebase context pre-loaded automatically.
Your reading, made intelligent.
Save any web page in one click. The more you save, the smarter it gets — Clip builds a personal knowledge base from your reading habit, automatically.
Why Pyckle
Pyckle = preservation. In a world where AI systems constantly reset and forget, we're the constant.
We ensure your most valuable asset—your context—remains accessible, optimized, and alive. Whether you're using Claude, GPT, Gemini, or whatever comes next.
Works with Claude Code, Cursor, Windsurf, and any MCP-compatible client.