Skip to main content

ANIX Code

tests python license status

🇬🇧 English · 🇷🇺 Русский

⚠️ Early version. The main things work, but bugs are possible: please report them in Issues.

An AI coding agent for your terminal. Paste an API key from any provider, describe a task, and the agent reads the project, edits files and runs commands by itself. Every change is shown as a diff, and nothing is changed without your approval (unless you turn on full access).

◆ ANIX Code  ~/projects/my-app                                        BUILD
 ▎ add a /health endpoint to api.py with a test
   ✓ → read api.py  84 lines
   ✓ ⌕ grep def test_  6 matches
   ✓ ✎ edit api.py  +7 −0
   ✓ $ pytest -q  → 0
 Done: added GET /health and a test in tests/test_api.py, all 7 tests pass.

Install

Python 3.11+ is required.

pipx install git+https://github.com/an1zy/anixcode     # one command
# or for development:
git clone https://github.com/an1zy/anixcode && cd anixcode
pip install -e .
anixcode            # open in the current folder
anixcode ~/proj     # or in a specific one

On the first start a connection window opens: choose a provider, paste the key and press “Model list”. Any OpenAI-compatible server works:

Provider Key
DeepSeek platform.deepseek.com
OpenAI, Anthropic, Google Gemini, Groq in their consoles
ANIX Router, Ollama, LM Studio not needed, they run locally
★ Your own providers “+ Own” button: name, /v1 address, key, “Check”

The model has to support tool calling (function calling).

If you start ANIX Code from your home folder or C:\Windows\System32 (a fresh cmd window), it asks which project to work in. Switch projects any time with Ctrl+P or by clicking the path in the header; you can also paste a https://github.com/... link there to clone a repository.

The interface language follows your system (Russian or English). Switch it with /language.

Usage

Keys
Enter / Ctrl+J send / new line
Tab / Shift+Tab Build (edits code) ↔ Plan (only reads and plans)
Esc stop the agent
Ctrl+P choose a project: recent, found git repositories, browse disks, clone by link
F2 full access on/off
Ctrl+O model and key
Ctrl+N / Ctrl+S new session / past sessions
Ctrl+T theme picker with preview: ANIX (peach), Noir (black & white), Tokyo Night, Catppuccin, Dracula, Nord…
F1 help

Click an action card to expand its diff or command output. Logo and progress animations can be turned off in settings.json: "animations": false.

Approvals. When the agent wants to change a file or run a command, a line appears above the input (the diff or the command is shown in the card in the chat): Enter allows, Esc denies, typing text and pressing Enter denies and tells the agent what to do instead. Dangerous commands (rm -rf, Remove-Item -Recurse, del /s, git push --force, git reset --hard…) are highlighted in red. Full access (F2 or click “🔒 ask first” at the bottom right): the agent works without asking until you turn it off.

@files. Type @ in a prompt to get suggestions of project files (↑↓, Enter or Tab). The whole file is sent with the prompt and the agent can edit it right away: fix the bug in @src/api.py. @folder/ attaches a listing of the folder.

Project rules. /init: the agent studies the project and writes AGENTS.md (what the project is, structure, build and test commands, code style). Add your own rules there, for example “tests use vitest, don't touch the legacy folder”. The agent reads this file before every task.

Reliability:

  • /undo reverts all file changes from the agent's last answer (several times in a row, also after a restart); /redo brings them back. Terminal commands (bash) are not undone.
  • The bottom right shows how full the model context is. At 80% the history is compressed automatically: old command outputs are removed first, then old steps are folded into a summary. Manually: /compact.
  • On network failures, timeouts, 429 and 5xx responses the request is retried up to 4 times.
  • The context size is guessed from the model name. For your own model set it in settings.json: "context_window": 128000.

Commands

Type / to get the list with descriptions (↑↓, Tab completes, Enter runs). The set follows Claude Code and OpenCode.

Command
/init study the project and write AGENTS.md
/review [focus] review uncommitted changes (or the last commit)
/commit [details] commit with a good message in the project's style
/test [what] find and run the tests, fix failing ones
/fix <error> fix an error from its text or a description
/explain <@file> explain code
/pr pull request description for the current branch
/undo / /redo undo the last answer's changes / bring them back
/compact compress the history
/new, /clear, /sessions, /resume sessions
/model, /connect model and key
/project choose a project
/theme, /mode, /permissions theme, Build/Plan, full access
/status summary: project, model, tokens, context
/export save the session as Markdown
/doctor check git, ripgrep, connection to the model, AI-RPC
/language interface language: Русский / English (after a restart)

Quick input:

  • !command runs a command in the terminal yourself, without the model; the agent sees the output: !pytest -q, then “fix it”.
  • # text saves a rule to AGENTS.md: # use pnpm, not npm.

Own commands. A Markdown file in .anixcode/commands/name.md (this project) or in %APPDATA%\anixcode\commands\name.md / ~/.config/anixcode/commands/name.md (all projects) becomes the command /name. $ARGUMENTS is replaced with everything after the command, $1, $2 with single words. An example is in .anixcode/commands/check.md:

---
description: run the project tests and show the result
---
Run `pytest -q`. If tests fail, show the failing test names and the short reason for each. $ARGUMENTS

Prompt commands also work without the interface: anixcode run "/review".

Without the interface:

anixcode run "find where the discount is calculated and explain it" --plan
anixcode run "fix the failing test" -y      # -y: allow everything without asking
anixcode connect                             # connect a model from the console
anixcode models                              # list the provider's models

Tools

Tool
read reads a file with line numbers, big files in parts
ls, glob folder tree, find files by pattern
grep search in code (ripgrep if installed)
edit precise text replacement; the file must have been read and not changed since
write new file or full rewrite
bash commands: tests, git, builds (PowerShell on Windows)

File tools never leave the project folder. Edits and commands need approval unless full access is on. Fine-tuning in settings.json: "permissions": {"edit": "ask", "exec": "ask"} (ask / allow / deny).

AI-RPC. If AI-RPC is running, Discord shows “ANIX Code · Generating”.

Updates and keys

Keys and settings are stored outside the code folder, so you can unpack a new version over the old one or into another folder: keys, own providers and sessions stay. To move to another computer, copy the whole folder below.

%APPDATA%\anixcode\ (Windows) or ~/.config/anixcode/:

  • settings.json: provider, model, theme, permissions, language;
  • auth.json: API keys (only on your computer, readable only by your user);
  • sessions.db: session history.

Development

pip install -e ".[dev]" && pytest

Full self-test: python selftest.py (on Windows you can double-click selftest.bat). It runs the unit tests, then starts the interface invisibly with a fake model and checks the answers, approvals, full access, /undo and /redo, / and @ suggestions, !command, # note, Build/Plan, /status, /doctor, /export, /compact, all windows, stopping with Esc and /new. No API key needed, no money spent, your settings and sessions are not touched.

Translations: interface strings are written in Russian in the code as tr("…"); English versions live in anixcode/i18n_en.py (a test checks that every string has one).

anixcode/
  llm.py          OpenAI-compatible API client: streaming, tool calls, readable errors, retries
  agent.py        the agent loop, approvals, Build/Plan modes, undo/redo
  tools/          read, ls, glob, grep, edit, write, bash
  prompts.py      system prompt + AGENTS.md
  storage.py      sessions and file snapshots in SQLite
  checkpoints.py  file snapshots for /undo
  compaction.py   history compression to fit the context
  danger.py       spotting dangerous commands
  mentions.py     @files: suggestions and attaching
  commands.py     /commands: built-in, ready-made prompts, own ones from .md
  i18n.py         Russian / English interface
  providers.py    provider list
  tui/            the Textual interface

Roadmap

  • Undo (/undo, /redo)
  • History compression, retries on failures
  • @file in prompts, /init, dangerous-command highlighting
  • /review, /commit, /test…, own commands from .md, ! and #
  • English interface
  • MCP servers
  • Several agents in parallel
  • Desktop version

Metadata

Release files for anixcode 0.8.0

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for anixcode 0.8.0
File Size Uploaded Provenance Actions
anixcode-0.8.0.tar.gz 91.1 kB Without provenance Details for anixcode-0.8.0.tar.gz

Built distribution (wheel)

Table of built distributions (wheels) for anixcode 0.8.0
File Interpreter ABI Platform Actions
anixcode-0.8.0-py3-none-any.whl Python 3 none any Details for anixcode-0.8.0-py3-none-any.whl

Total release size: 182.7 kB

Release files / anixcode-0.8.0.tar.gz

Download URL anixcode-0.8.0.tar.gz
Size 91.1 kB
Tags Source
SHA-256 checksum
How to use checksums
add7f2959a799fdc48cb0249ad040e4bf52f9a021c4f156d7660dd01d8c59389
BLAKE2b-256 checksum
How to use checksums
ef417ffe2344115ff5a6acd5e5a8f87b9b7a2dcf8be300bffc6d247ea87826fe
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.14.3

Release files / anixcode-0.8.0-py3-none-any.whl

Download URL anixcode-0.8.0-py3-none-any.whl
Size 91.6 kB
Tags Python 3
SHA-256 checksum
How to use checksums
4039c2ad1343da7be42fe3326a7265a698f3369df4c6fa0d11ad47584b00da7c
BLAKE2b-256 checksum
How to use checksums
970d197152e4c52e365a6623d6aa6a361ca7dfbb410b6231f3333158c5eff5f3
Upload date
Uploaded using Trusted Publishing?
What is trusted publishing?
No
Uploaded via twine/7.0.0 CPython/3.14.3

Security for anixcode 0.8.0

Provenance

Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.

No provenance found

None of the files in this release were uploaded with provenance metadata.

A note on package security

PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.

To learn more about protecting yourself from malicious actors, visit our package security documentation.

Release history Release notifications | RSS feed

This release

0.8.0 This release

2 release files No provenance

Anthropic, PBC Visionary sponsor Bloomberg Visionary sponsor Hudson River Trading Visionary sponsor Meta Visionary sponsor NVIDIA Visionary sponsor Microsoft Sustainability sponsor Depot Continuous Integration AWS Cloud computing and Security Sponsor Datadog Monitoring Fastly CDN Google Download Analytics Sentry Error logging StatusPage Status page