● ai-creed

← projects

ai-ezio

[public]

hax-derived, workflow-native coding agent and generic mcp host for the ai-* ecosystem

v0.2.0-beta.0 — early beta, on npm.

what it does

ai-ezio is a coding agent derived from hax — kept fast and small — wrapped in a TypeScript harness that makes it machine-controllable. It keeps hax’s C engine headless and speaks an explicit protocol over file descriptors, so the rest of the ecosystem can drive it without scraping a terminal UI. It is also a generic MCP host: the model can call any configured MCP server mid-turn.

why

The ai-* workflows need an agent they can mount, drive, and trust — one with explicit turn boundaries and handback text instead of fragile TUI parsing. ezio keeps hax’s strengths as a terminal coding assistant while adding the workflow-native behavior mounted collaboration requires.

# features

  • fast, small terminal coding agent — hax's c engine kept headless
  • explicit machine protocol over fds — no tui scraping for mounted automation
  • generic mcp host — the model calls any configured stdio mcp server live, mid-turn
  • first-class ai-whisper agent — mounts as implementer or reviewer in workflows
  • context compaction and session recording built in
  • one install, hax binary embedded — macos and linux via npm prebuilts

the hybrid

hax      = c engine — streaming, providers, tool calls
ai-ezio  = ts harness — protocol, mcp host, session recording, cli

The C engine stays minimal and fast; the TypeScript side owns the protocol, the MCP host, and the integration contract. ai-ezio is a downstream fork: hax is created and authored by Oleksandr Chekhovskyi — its authorship and lineage are kept explicit in the project’s NOTICE and UPSTREAM.md.

mcp host

ezio is a generic MCP host. The model calls configured stdio MCP servers live, mid-turn, under an allow / deny / confirm policy, with tools namespaced per server. ai-cortex is the first-class memory and context server it pairs with.

in the ecosystem

  • ai-whisper — ezio is a first-class mountable agent type; it can take the implementer or reviewer role in spec-driven-development, ralph-loop, and complex-bug-fixing runs.
  • ai-cortex — mounted as an MCP server for project context and memory.

# install

npm i -g ai-ezio