GitHub Repository

Three packages: @kernel.chat/agent-os (POSIX for AI agents — capabilities, namespaces, quotas, taint, audit, vault, outcomes), @kernel.chat/kbot (terminal AI agent, MCP-native, BYOK), @kernel.chat/kbot-finance (audit-grade AI for regulated industries). Provenance-engineering substrate.

15 starsTypeScript

Kbot – AI agent that forges its own tools at runtime (MIT, $0)

Name: Kbot – AI agent that forges its own tools at runtime (MIT, $0)
Availability: InStock
Author: isaacsight

by isaacsight·Mar 26, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidWizardryBold Bet

Terminal agent that writes persistent CLI tools when it hits capability gaps.

Strengths

•HMAC memory integrity prevents prompt injection, supports 20 LLM providers.
•Bayesian skill routing adapts to user patterns over time without cloud.

Weaknesses

•Terminal agent market is saturated, GitHub stats show minimal adoption.
•Marketing claims need verification against established tools like Aider.

Post Description

I built kbot — a terminal AI agent that learns from every session and forges new tools when it hits problems it can't solve.

What makes it different: Self-improving (Bayesian skill routing gets smarter over time). Tool forging (encounters a gap, builds a tool, tool persists forever). Self-defense (HMAC memory integrity, prompt injection detection). 368 tools, 41 agents, 19 local models. Fully offline, $0. Dream mode (improves itself while idle). MIT licensed.

Demo of kbot forging a tool in 30 seconds: https://raw.githubusercontent.com/isaacsight/kernel/main/too...

npm install -g @kernel.chat/kbot