Aivolut
CogniKernel screenshot

CogniKernel

Local, persistent memory for AI coding-agent sessions.

What is CogniKernel?

CogniKernel is local memory software for AI coding agents that preserves useful project context between sessions. It captures decisions, constraints, and abandoned approaches, then adds a compact context block when work resumes.

The project treats extraction as classification rather than generative summarization. Its documented local pipeline uses small ONNX encoder models to score salience and detect superseded decisions, without sending session content to a cloud service.

Developers can use it to share one project-path-keyed store across Claude Code and Codex. The tool includes automatic capture and recall, MCP tools for targeted retrieval, and a CLI for setup and maintenance.

CogniKernel is aimed at developers who want coding agents to retain project-specific decisions without continually rebuilding context. Its hooks can surface relevant memory at session start, while prompting, before tools run, and when a session ends.

The software is installed from PyPI and runs locally, making it a fit for teams that prefer to keep coding-session memory on their own machines. Pricing was not verified from the official repository.

CogniKernel Features

Persistent local project memory

CogniKernel watches coding sessions, extracts decisions and constraints worth retaining, and stores them for a later session. Its memory store is keyed to a project path so the same context can move between supported coding tools in one directory.

Local classification instead of cloud summarization

The project documents a deterministic sanitize, classify, and consolidate pipeline for identifying useful session memory. Its stated design uses local fine-tuned encoder models and does not require an API key or a separate generative model for extraction.

Hooks for capture, recall, and guardrails

CogniKernel attaches to four points in a coding-agent session, including startup, prompting, tool use, and stopping. This lets it inject a session block, surface relevant memories, warn on past decisions, and persist new decisions after a session.

Hybrid retrieval and typed memories

The tool stores typed decisions, constraints, and abandoned approaches instead of only free-text chunks. Its retrieval design combines lexical search with optional dense signals and gives special handling to rules that prohibit a past approach.

Integrations

Claude CodeCodex

Pricing

Pricing not verified

Coding & Developer ToolsAI coding agentsDeveloper memoryLocal AIMCPClaude CodeCodex

User reviews

No reviews yet. Be the first to review this tool.

Log in to write a review.

← Back to AI Tools