DEV Community

Cover image for Announcing ÆtherCodex — a Hermetic AI Coding Oracle (TextMate today, portable core)
dantiel
dantiel

Posted on

Announcing ÆtherCodex — a Hermetic AI Coding Oracle (TextMate today, portable core)

ÆtherCodex is an open-source coding oracle that treats software work as a deliberate sequence of transformations — not one-off prompts. It currently ships as a TextMate bundle, but its core “verbs” are editor-agnostic and can be adapted to VS Code, Neovim, JetBrains, etc. GitHub

Why a “Magnum Opus”?

At the center is the Magnum Opus Engine — a 10-step pipeline for multi-phase tasks:
Nigredo → Albedo → Citrinitas → Rubedo → Solve → Coagula → Test → Purificatio → Validatio → Documentatio.

Each step carries forward context, results, and guidance, so changes, tests, and decisions cohere into a single flow with persistence. GitHub

What can it do?

ÆtherCodex exposes a small set of durable, composable actions that map cleanly to any editor:

  • read_file / patch_file / create_file — precise diffs & edits (with AST-grep semantics available)
  • run_command — shell rituals for builds/tests/scripts
  • remember / recall — a memory palace for context continuity
  • create_task / execute_task — orchestrate multi-step work via the Opus
  • file_overview — structural hints (classes, methods, line targets) without full-file reads These “verbs” are the portable core; TextMate is just the first host. dantiel.github.io

TextMate today

The repo contains a standard TextMate bundle (info.plist, .tm_properties). Install by cloning and placing it in your TextMate Bundles directory; the docs page shows the basic “Summoning the Oracle” steps. GitHub

git clone https://github.com/dantiel/aethercodex.git
cd aethercodex
# install into your TextMate Bundles directory (see TextMate docs / your setup)
Enter fullscreen mode Exit fullscreen mode

Porting to other editors (how)

Because the core is expressed as verbs and a stateful task engine, porting is mostly adapter work:

  1. Command binding: map editor events/commands to read_file, patch_file, run_command, etc.
  2. Buffer & fs access: implement getters/setters that the engine calls for file I/O and diffs.
  3. Task UI: surface Opus step status, logs, and errors in the host editor’s panel/toasts.
  4. Memory integration: persist/reload “Mnemosyne” notes (JSON/kv/db — your choice).
  5. Symbolic overview: reuse the Argonaut/overview call to give “jump-to-line” navigation hints. The repo & site outline these components (Mnemosyne/Aegis, Argonaut, Pythia UI, Hermetic Command Synthesis). 

Why this paradigm is useful

  • Deterministic shape: multi-step tasks are explicit and replayable.
  • Context continuity: results persist across steps (and sessions).
  • Editor-agnostic: a handful of verbs keeps the surface area small while staying powerful.
  • Token-aware: symbolic overview + selective reads reduce LLM context bloat. 

Roadmap highlights

Future features on the site include deeper symbolic analysis, multi-file “astral projection,” ritual command sequences, divination/forecasting modes, and secure code oracles. PRs welcome if you want to help shape these. 


Links

If this paradigm resonates with you — whether you’re into TextMate, Neovim, or building editor agents — I’d love your feedback, adapters, and experiments.

Solve. Coagula. Purificatio. Documentatio.

Top comments (0)