refactor: move OpenCode artifacts under opencode/

arka/public is the dump for all Arka downloads. Keep a root stub so the
old installer URL still works. Document in-OpenCode bash install.
This commit is contained in:
ixoblakp committed 2026-08-27 17:56:04 +03:00
1 parent 6e0dc4d02b
commit 22938b01ca
9 files changed
+585 -490

No files matched your search

+26 -178
View File
@@ -1,187 +1,35 @@
# OpenCode и модели Arka AI # arka/public
Arka AI — OpenAI-совместимый API (`https://api.arka-ai.ru/v1`). В каталоге models.dev его нет, поэтому в свежем OpenCode провайдера **Arka-AI** и списка моделей нет, пока не поставить плагин. Публичные файлы Arka AI: загрузки, инструкции, скрипты. Анонимный raw, без токена.
Одна команда ставит плагин, skill и (по желанию) ключ. Каталог чат-моделей подтягивается сам (~116 из ~216; без TTS/картинок/видео/эмбеддингов). Снимок каталога обновляется каждый день в **05:00 по Москве**. `https://git.arka-ai.ru/arka/public/raw/branch/main/<path>`
Нужен [API-ключ Arka AI](https://arka-ai.ru/docs/account/api-keys). Секрет ключа показывается **один раз** при создании. Каждый продукт — своя папка на `main`. Приватный код сюда не кладём.
| Папка | Что |
|-------|-----|
| [`opencode/`](opencode/) | Провайдер OpenCode **Arka-AI**: установщик, плагин, skill, каталог чат-моделей |
Однострочник OpenCode:
```bash
curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/install-arka-provider.sh | bash
```
--- ---
## Быстрый старт Public Arka AI downloads, docs, and scripts. Anonymous raw URLs, no token.
`https://git.arka-ai.ru/arka/public/raw/branch/main/<path>`
One top-level folder per product. No private source.
| Folder | What |
|--------|------|
| [`opencode/`](opencode/) | OpenCode **Arka-AI** provider: installer, plugin, skill, chat catalog |
OpenCode one-liner:
```bash ```bash
curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/install-arka-provider.sh | bash curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/install-arka-provider.sh | bash
``` ```
Скрипт спросит `ARKA_API_KEY` (можно оставить пустым и задать позже).
Ключ и модель по умолчанию сразу:
```bash
curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/install-arka-provider.sh \
| bash -s -- --key "$ARKA_API_KEY" --set-default DeepSeek-V4-Pro
```
Дальше:
1. Откройте **новый терминал** (чтобы подхватился `export ARKA_API_KEY`).
2. Перезапустите OpenCode.
3. Команда `/models` — в списке должен быть провайдер **Arka-AI** и чат-модели.
Если OpenCode запущен из GUI / иконки / сокета, а не из login-shell, переменная из `~/.bashrc` до него может не дойти. Задайте `ARKA_API_KEY` в том окружении, откуда стартует OpenCode, или введите ключ в интерактивном приглашении установщика.
### Что ставится
| Куда | Что |
|------|-----|
| `~/.config/opencode/plugins/arka-ai.ts` | Плагин провайдера **Arka-AI** |
| `~/.config/opencode/skills/arka-provider/SKILL.md` | Skill: как обновить каталог и сменить модель |
| `~/.bashrc` (или первый из `.zshrc` / `.profile`) | `export ARKA_API_KEY=…`, если ключ передали |
| `~/.config/opencode/opencode.json` | Только при `--set-default`: `"model": "Arka-AI/<id>"`. Существующий конфиг **не затирается**. |
Установщик **офлайн** и идемпотентный: сеть не трогает, повторный запуск перезаписывает плагин и skill каноническими копиями.
`python3` нужен только с `--set-default`. Отдельный `bun` на машине не нужен — у OpenCode свой runtime.
### Модель по умолчанию
Идентификаторы — как в каталоге, с префиксом провайдера:
```json
{
"model": "Arka-AI/DeepSeek-V4-Pro"
}
```
Примеры: `Arka-AI/DeepSeek-V4-Pro`, `Arka-AI/qwen3.8-max`, `Arka-AI/kimi-k3`, `Arka-AI/MiniMax-M3`, `Arka-AI/GLM-5.3`, `Arka-AI/deepseek-v4-flash-0731`.
Их всегда оставляем в списке, даже если фильтр чат-моделей их бы отсёк.
### Обновить каталог
```bash
rm -f ~/.cache/opencode/arka-ai-models.json
```
Перезапустите OpenCode. Плагин сначала берёт публичный снимок
`https://git.arka-ai.ru/arka/public/raw/branch/main/catalog.json`
(без ключа), при ошибке — живой `https://api.arka-ai.ru/v1/models`. Локальный кэш живёт 24 часа. Конфиг OpenCode на лету не перечитывается.
Список моделей (ключ не нужен):
```bash
curl -sS https://git.arka-ai.ru/arka/public/raw/branch/main/catalog.json | python3 -m json.tool
```
Вызовы моделей идут на `https://api.arka-ai.ru/v1` с `ARKA_API_KEY`.
### Если что-то не так
| Симптом | Что проверить |
|---------|----------------|
| После `/models` нет Arka-AI | Есть ли `~/.config/opencode/plugins/arka-ai.ts`; экспортирован ли `ARKA_API_KEY` **в процессе OpenCode**; открыт ли новый шелл |
| Каталог не меняется | `rm -f ~/.cache/opencode/arka-ai-models.json` и рестарт |
| Ошибка импорта плагина | Снова запустить установщик — он вернёт канонический файл |
| `curl` отдаёт HTML, не скрипт | Нужен путь `/raw/branch/main/…`, не страница репозитория |
---
# OpenCode and Arka AI models
Arka AI is an OpenAI-compatible API at `https://api.arka-ai.ru/v1`. It is not in models.dev, so a fresh OpenCode install has no **Arka-AI** provider until you add the plugin.
One command installs the plugin, a companion skill, and (optionally) your API key. The chat catalog loads on OpenCode start (~116 of ~216 models; TTS, image, video, and embeddings are filtered out). The published catalog snapshot refreshes daily at **05:00 Europe/Moscow**.
You need an [Arka AI API key](https://arka-ai.ru/docs/account/api-keys). The secret is shown **once** at creation.
## Quick start
```bash
curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/install-arka-provider.sh | bash
```
The script prompts for `ARKA_API_KEY` (blank is allowed; set it later).
Key and default model in one go:
```bash
curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/install-arka-provider.sh \
| bash -s -- --key "$ARKA_API_KEY" --set-default DeepSeek-V4-Pro
```
Then:
1. Open a **new shell** so `export ARKA_API_KEY` is picked up.
2. Restart OpenCode.
3. Run `/models` — you should see provider **Arka-AI** and the chat models.
If OpenCode is started from a GUI, dock icon, or socket rather than a login shell, `~/.bashrc` may not reach it. Set `ARKA_API_KEY` in that process environment, or type it at the installer prompt.
### What the installer writes
| Path | Role |
|------|------|
| `~/.config/opencode/plugins/arka-ai.ts` | **Arka-AI** provider plugin |
| `~/.config/opencode/skills/arka-provider/SKILL.md` | How to refresh the catalog and change the default model |
| `~/.bashrc` (else first existing `.zshrc` / `.profile`) | `export ARKA_API_KEY=…` when a key is given |
| `~/.config/opencode/opencode.json` | Only with `--set-default`: `"model": "Arka-AI/<id>"`. An existing config is **not** replaced. |
The installer is **offline** and idempotent: no network calls; re-running overwrites plugin and skill with the canonical copies.
`python3` is required **only** with `--set-default`. You do not need `bun` on the host — OpenCode ships its own runtime.
### Default model
Use catalog ids with the provider prefix:
```json
{
"model": "Arka-AI/DeepSeek-V4-Pro"
}
```
Examples: `Arka-AI/DeepSeek-V4-Pro`, `Arka-AI/qwen3.8-max`, `Arka-AI/kimi-k3`, `Arka-AI/MiniMax-M3`, `Arka-AI/GLM-5.3`, `Arka-AI/deepseek-v4-flash-0731`.
Those ids stay visible even if the chat filter would drop them.
### Refresh the catalog
```bash
rm -f ~/.cache/opencode/arka-ai-models.json
```
Restart OpenCode. The plugin fetches the public snapshot
`https://git.arka-ai.ru/arka/public/raw/branch/main/catalog.json`
(no key), then falls back to live `https://api.arka-ai.ru/v1/models`. Local cache TTL is 24 hours. OpenCode does not hot-reload config.
Inspect the published list (no key):
```bash
curl -sS https://git.arka-ai.ru/arka/public/raw/branch/main/catalog.json | python3 -m json.tool
```
Inference still uses `ARKA_API_KEY` against `https://api.arka-ai.ru/v1`.
### Troubleshooting
| Symptom | Check |
|---------|--------|
| No Arka-AI after `/models` | `~/.config/opencode/plugins/arka-ai.ts` exists; `ARKA_API_KEY` is set **in the OpenCode process**; you opened a new shell |
| Models look stale | `rm -f ~/.cache/opencode/arka-ai-models.json` and restart |
| Plugin import error | Re-run the installer to restore the canonical file |
| `curl` returns HTML | Use `/raw/branch/main/…`, not the repo page |
---
## This repository
Public, anonymous files. Private product code lives in other Gitea repos.
| File | Role |
|------|------|
| `install-arka-provider.sh` | Self-contained installer (the one-liner) |
| `plugins/arka-ai.ts` | Plugin source (embedded in the installer) |
| `skills/arka-provider/SKILL.md` | Skill source (embedded in the installer) |
| `catalog.json` | Filtered chat catalog, refreshed daily at 05:00 Europe/Moscow |
| `README.md` | This page — copy the Russian or English section into arka-ai.ru «агентные помощники» / agentic helpers |
+6 -302
View File
@@ -1,304 +1,8 @@
#!/usr/bin/env bash #!/usr/bin/env bash
# # Compatibility stub. Canonical installer:
# install-arka-provider.sh — one-shot installer for the arka-ai.ru LLM provider # https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/install-arka-provider.sh
# in a fresh opencode install.
#
# It is fully self-contained: it embeds the provider plugin and the companion
# skill, writes them into ~/.config/opencode, and persists ARKA_API_KEY.
#
# Usage:
# bash install-arka-provider.sh # prompt for API key
# ARKA_API_KEY=xxx bash install-arka-provider.sh # key from env (no prompt)
# bash install-arka-provider.sh --key xxx --set-default DeepSeek-V4-Pro
#
# Options:
# --key KEY ARKA_API_KEY to persist (default: $ARKA_API_KEY, else prompt)
# --set-default MODEL also set "model": "Arka-AI/<MODEL>" in opencode.json
# -h, --help show this help
#
# Idempotent: safe to run multiple times. Overwrites plugin/skill with the
# embedded versions; never clobbers an existing opencode.json (only upserts the
# requested default model). Makes no network calls.
set -euo pipefail set -euo pipefail
echo "==> installer lives at opencode/install-arka-provider.sh" >&2
usage() { curl -fsSL \
sed -n '2,21p' "$0" 'https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/install-arka-provider.sh' \
} | bash -s -- "$@"
KEY=""
MODEL=""
while [ $# -gt 0 ]; do
case "$1" in
--key) KEY="$2"; shift 2 ;;
--key=*) KEY="${1#*=}"; shift ;;
--set-default) MODEL="$2"; shift 2 ;;
--set-default=*) MODEL="${1#*=}"; shift ;;
-h|--help) usage; exit 0 ;;
*) printf 'unknown option: %s\n' "$1" >&2; usage; exit 1 ;;
esac
done
CONF_DIR="${XDG_CONFIG_HOME:-$HOME/.config}/opencode"
PLUGIN_DIR="$CONF_DIR/plugins"
SKILL_DIR="$CONF_DIR/skills/arka-provider"
mkdir -p "$PLUGIN_DIR" "$SKILL_DIR" "$HOME"
echo "==> writing provider plugin to $PLUGIN_DIR/arka-ai.ts"
cat > "$PLUGIN_DIR/arka-ai.ts" <<'PLUGIN'
import { homedir } from "node:os"
import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs"
import { join } from "node:path"
const PROVIDER_ID = "Arka-AI"
const API_URL = "https://api.arka-ai.ru/v1"
const CATALOG_URL =
"https://git.arka-ai.ru/arka/public/raw/branch/main/catalog.json"
const CACHE_FILE = join(homedir(), ".cache", "opencode", "arka-ai-models.json")
const TTL_MS = 24 * 60 * 60 * 1000
// Ids that must always be exposed, even if the filter would drop them.
const FORCED_MODELS = [
"DeepSeek-V4-Pro",
"deepseek-v4-flash-0731",
"qwen3.8-max",
"kimi-k3",
"MiniMax-M3",
"GLM-5.3",
]
// Heuristic: drop non-chat models (TTS/ASR/image/video/OCR/embeddings/rerankers…)
// from the catalog so they don't clutter the /models picker.
// Keep in sync with scripts/sync-catalog.py
const NON_CHAT = new RegExp(
[
"tts", "asr", "audio", "speech", "voice", "whisper", "sensevoice",
"image", "(^|[.-])img", "wan", "vidu", "video", "sora", "flux",
"diffus", "kolors", "cogview", "ocr", "embedding", "(^|[.-])embed",
"rerank", "clip", "moderation", "(^|[.-])filter", "detect", "classif",
"(^|[.-])location", "(^|[.-])search", "comfyui", "meme", "hello",
"(^|[.-])sam", "florence", "esrgan", "rmbg", "background", "poster",
"(^|[.-])face", "(^|[.-])lip", "translate", "(^|[.-])mt([2-]|$)",
"(^|[.-])bge", "(^|[.-])e5", "(^|[.-])jina", "(^|[.-])mpnet",
"spark-tts", "cosyvoice", "chattts", "indextts", "step-audio",
"fish-speech", "cosy", "mate-gguf", "tokenizer",
].join("|"),
"i",
)
type ModelMap = Record<string, { name: string }>
type Cache = { ts: number; models: ModelMap }
function loadCache(): Cache | null {
try {
if (existsSync(CACHE_FILE)) return JSON.parse(readFileSync(CACHE_FILE, "utf8"))
} catch {}
return null
}
function saveCache(models: ModelMap): void {
try {
mkdirSync(join(homedir(), ".cache", "opencode"), { recursive: true })
writeFileSync(CACHE_FILE, JSON.stringify({ ts: Date.now(), models }))
} catch {}
}
async function fetchJson(
url: string,
headers: Record<string, string>,
timeoutMs: number,
): Promise<unknown> {
const ctrl = new AbortController()
const timer = setTimeout(() => ctrl.abort(), timeoutMs)
try {
const res = await fetch(url, { headers, signal: ctrl.signal })
if (!res.ok) throw new Error(`HTTP ${res.status}`)
return await res.json()
} finally {
clearTimeout(timer)
}
}
function idsFromPayload(json: unknown): string[] {
const data = (json as { data?: Array<{ id: string }> }).data
return (data ?? []).map((m) => m.id).filter(Boolean)
}
function toModelMap(ids: string[], filter: boolean): ModelMap {
const models: ModelMap = {}
for (const id of ids) {
if (!filter || FORCED_MODELS.includes(id) || !NON_CHAT.test(id)) {
models[id] = { name: id }
}
}
return models
}
async function resolveModels(): Promise<ModelMap | null> {
const cache = loadCache()
if (cache && Date.now() - cache.ts < TTL_MS) return cache.models
try {
const models = toModelMap(idsFromPayload(await fetchJson(CATALOG_URL, {}, 5000)), false)
if (Object.keys(models).length) {
saveCache(models)
return models
}
} catch {}
try {
const key = process.env.ARKA_API_KEY
const json = await fetchJson(
`${API_URL}/models`,
key ? { Authorization: `Bearer ${key}` } : {},
5000,
)
const models = toModelMap(idsFromPayload(json), true)
saveCache(models)
return models
} catch {
return cache?.models ?? null
}
}
export default async function ArkaAIProviderPlugin() {
const models = await resolveModels()
return {
config: (cfg: any) => {
cfg.provider ??= {}
const prev = cfg.provider[PROVIDER_ID] ?? {}
// Static models from opencode.json act as a fallback; the fetched
// catalog is layered on top so the configured default model never
// disappears even if the catalog drops it.
cfg.provider[PROVIDER_ID] = {
name: "Arka-AI",
api: API_URL,
env: ["ARKA_API_KEY"],
...prev,
models: { ...(prev.models ?? {}), ...(models ?? {}) },
}
},
}
}
PLUGIN
echo "==> writing companion skill to $SKILL_DIR/SKILL.md"
cat > "$SKILL_DIR/SKILL.md" <<'SKILL'
---
name: arka-provider
description: Manage the custom Arka-AI (arka-ai.ru) LLM provider in opencode config. Use when the user needs to add, refresh, or fix the arka-ai.ru provider, sync its model catalog, change the Arka-AI default model, or when arka models are missing from /models.
---
# Arka-AI provider
arka-ai.ru is **not** in models.dev, so opencode ships no built-in config for it.
It is wired up locally with a plugin + a static `provider` block in
`~/.config/opencode/opencode.json`.
## How it works
- Plugin: `~/.config/opencode/plugins/arka-ai.ts`
- On startup it loads the published chat catalog
`https://git.arka-ai.ru/arka/public/raw/branch/main/catalog.json`
(no API key; refreshed daily at 05:00 Europe/Moscow).
- If that fetch fails, it falls back to `https://api.arka-ai.ru/v1/models`
and filters out non-chat models (TTS/ASR/image/video/OCR/embedding/reranker/…)
via a heuristic regex so the `/models` picker stays clean.
- Caches the result at `~/.cache/opencode/arka-ai-models.json` (TTL 24h).
- Injects the provider into config via the `config` hook, layered over the
static block in `opencode.json`.
- Static block (`opencode.json` → `provider["Arka-AI"]`) holds
`name`, `api`, `env`, and a small fallback `models` map that is used when the
catalog is unreachable.
- Inference still uses `ARKA_API_KEY` against `https://api.arka-ai.ru/v1`.
## Refresh the model list
Catalog ~216 raw models / ~116 chat; the public snapshot updates at 05:00
Europe/Moscow. Changes land in opencode only after cache expiry + restart
(config is not hot-reloaded).
1. `rm -f ~/.cache/opencode/arka-ai-models.json`
2. Restart opencode (or run `opencode` again) — the plugin re-fetches on the
next cold start.
Optionally inspect the live API (no key required for listing):
```bash
curl -sS https://api.arka-ai.ru/v1/models | python3 -m json.tool
curl -sS https://git.arka-ai.ru/arka/public/raw/branch/main/catalog.json | python3 -m json.tool
```
## Change default model
Edit `opencode.json` → `"model": "Arka-AI/<model-id>"`. Use the IDs exactly as
returned by the catalog (e.g. `Arka-AI/DeepSeek-V4-Pro`,
`Arka-AI/qwen3.8-max`, `Arka-AI/kimi-k3`).
## Adjust filtering
Edit `NON_CHAT` (regex) or `FORCED_MODELS` in the plugin file **and** the same
lists in `scripts/sync-catalog.py` in `arka/public`, then re-run the installer
and restart. `FORCED_MODELS` keeps must-have chat models even if the heuristic
would drop them.
SKILL
# --- API key -----------------------------------------------------------
if [ -z "$KEY" ]; then
read -rsp "ARKA_API_KEY (leave blank to skip): " KEY || true
echo
fi
if [ -n "$KEY" ]; then
RC=""
for f in "$HOME/.bashrc" "$HOME/.zshrc" "$HOME/.profile"; do
if [ -f "$f" ]; then RC="$f"; break; fi
done
RC="${RC:-$HOME/.bashrc}"
touch "$RC"
if grep -q 'ARKA_API_KEY' "$RC"; then
echo "==> ARKA_API_KEY already present in $RC (skipped)"
else
printf '\n# opencode Arka-AI provider\nexport ARKA_API_KEY="%s"\n' "$KEY" >> "$RC"
echo "==> wrote ARKA_API_KEY to $RC"
fi
export ARKA_API_KEY="$KEY"
else
echo "==> warning: no ARKA_API_KEY set; export it in your shell before starting opencode" >&2
fi
# --- default model (optional) ------------------------------------------
if [ -n "$MODEL" ]; then
CONF="$CONF_DIR/opencode.json"
echo "==> setting default model Arka-AI/$MODEL in $CONF"
python3 - "$CONF" "Arka-AI/$MODEL" <<'PY'
import json, os, sys
path, model = sys.argv[1], sys.argv[2]
data = {}
if os.path.exists(path):
with open(path) as f:
data = json.load(f)
data.setdefault("$schema", "https://opencode.ai/config.json")
data["model"] = model
with open(path, "w") as f:
json.dump(data, f, indent=2)
f.write("\n")
PY
fi
echo
echo "Done."
echo " - provider plugin: $PLUGIN_DIR/arka-ai.ts"
echo " - skill: $SKILL_DIR/SKILL.md"
echo
echo "Next steps:"
echo " 1. Restart opencode (open a new shell so ARKA_API_KEY is exported)."
echo " 2. Run /models — the Arka-AI catalog (~116 chat models) loads on first start."
echo " 3. Optional: refresh catalog later with: rm -f ~/.cache/opencode/arka-ai-models.json"
+237
View File
@@ -0,0 +1,237 @@
# OpenCode и модели Arka AI
Arka AI — OpenAI-совместимый API (`https://api.arka-ai.ru/v1`). В каталоге models.dev его нет, поэтому в свежем OpenCode провайдера **Arka-AI** и списка моделей нет, пока не поставить плагин.
Одна команда ставит плагин, skill и (по желанию) ключ. Каталог чат-моделей подтягивается сам (~116 из ~216; без TTS/картинок/видео/эмбеддингов). Снимок каталога обновляется каждый день в **05:00 по Москве**.
Нужен [API-ключ Arka AI](https://arka-ai.ru/docs/account/api-keys). Секрет ключа показывается **один раз** при создании.
---
## Быстрый старт
```bash
curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/install-arka-provider.sh | bash
```
Скрипт спросит `ARKA_API_KEY` (можно оставить пустым и задать позже).
Ключ и модель по умолчанию сразу:
```bash
curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/install-arka-provider.sh \
| bash -s -- --key "$ARKA_API_KEY" --set-default DeepSeek-V4-Pro
```
Дальше:
1. Откройте **новый терминал** (чтобы подхватился `export ARKA_API_KEY`).
2. Перезапустите OpenCode.
3. Команда `/models` — в списке должен быть провайдер **Arka-AI** и чат-модели.
Если OpenCode запущен из GUI / иконки / сокета, а не из login-shell, переменная из `~/.bashrc` до него может не дойти. Задайте `ARKA_API_KEY` в том окружении, откуда стартует OpenCode, или введите ключ в интерактивном приглашении установщика.
### Из уже открытого OpenCode
Отдельного магазина плагинов у OpenCode нет. Можно не выходить из CLI: попросите **текущую** модель (любую, не Arka) выполнить установщик через bash.
Вставьте в чат (подставьте свой ключ):
```
Поставь провайдер Arka AI для OpenCode. Не спрашивай ключ интерактивно.
Выполни одну команду и покажи вывод:
curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/install-arka-provider.sh | bash -s -- --key 'ARKA_API_KEY_HERE' --set-default DeepSeek-V4-Pro
После успеха напиши, что нужно полностью перезапустить OpenCode (конфиг не подхватывается на лету).
```
Что происходит:
1. Агент вызывает встроенный bash/shell.
2. OpenCode спросит разрешение на команду — нажмите Allow.
3. Скрипт пишет плагин, skill и ключ в `~/.config/opencode` (установщик сам по себе офлайн; `curl` только качает скрипт).
4. `--key` обязателен: приглашение «введите ключ» в неинтерактивном пайпе зависнет.
5. **Перезапустите OpenCode целиком** — плагин грузится только на старте. Затем `/models` → **Arka-AI**.
Пока сессия не перезапущена, разговор идёт на старой модели. Это нормально: установщик не требует Arka.
### Что ставится
| Куда | Что |
|------|-----|
| `~/.config/opencode/plugins/arka-ai.ts` | Плагин провайдера **Arka-AI** |
| `~/.config/opencode/skills/arka-provider/SKILL.md` | Skill: как обновить каталог и сменить модель |
| `~/.bashrc` (или первый из `.zshrc` / `.profile`) | `export ARKA_API_KEY=…`, если ключ передали |
| `~/.config/opencode/opencode.json` | Только при `--set-default`: `"model": "Arka-AI/<id>"`. Существующий конфиг **не затирается**. |
Установщик **офлайн** и идемпотентный: сеть не трогает, повторный запуск перезаписывает плагин и skill каноническими копиями.
`python3` нужен только с `--set-default`. Отдельный `bun` на машине не нужен — у OpenCode свой runtime.
### Модель по умолчанию
Идентификаторы — как в каталоге, с префиксом провайдера:
```json
{
"model": "Arka-AI/DeepSeek-V4-Pro"
}
```
Примеры: `Arka-AI/DeepSeek-V4-Pro`, `Arka-AI/qwen3.8-max`, `Arka-AI/kimi-k3`, `Arka-AI/MiniMax-M3`, `Arka-AI/GLM-5.3`, `Arka-AI/deepseek-v4-flash-0731`.
Их всегда оставляем в списке, даже если фильтр чат-моделей их бы отсёк.
### Обновить каталог
```bash
rm -f ~/.cache/opencode/arka-ai-models.json
```
Перезапустите OpenCode. Плагин сначала берёт публичный снимок
`https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/catalog.json`
(без ключа), при ошибке — живой `https://api.arka-ai.ru/v1/models`. Локальный кэш живёт 24 часа. Конфиг OpenCode на лету не перечитывается.
Список моделей (ключ не нужен):
```bash
curl -sS https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/catalog.json | python3 -m json.tool
```
Вызовы моделей идут на `https://api.arka-ai.ru/v1` с `ARKA_API_KEY`.
### Если что-то не так
| Симптом | Что проверить |
|---------|----------------|
| После `/models` нет Arka-AI | Есть ли `~/.config/opencode/plugins/arka-ai.ts`; экспортирован ли `ARKA_API_KEY` **в процессе OpenCode**; открыт ли новый шелл |
| Каталог не меняется | `rm -f ~/.cache/opencode/arka-ai-models.json` и рестарт |
| Ошибка импорта плагина | Снова запустить установщик — он вернёт канонический файл |
| `curl` отдаёт HTML, не скрипт | Нужен путь `/raw/branch/main/…`, не страница репозитория |
---
# OpenCode and Arka AI models
Arka AI is an OpenAI-compatible API at `https://api.arka-ai.ru/v1`. It is not in models.dev, so a fresh OpenCode install has no **Arka-AI** provider until you add the plugin.
One command installs the plugin, a companion skill, and (optionally) your API key. The chat catalog loads on OpenCode start (~116 of ~216 models; TTS, image, video, and embeddings are filtered out). The published catalog snapshot refreshes daily at **05:00 Europe/Moscow**.
You need an [Arka AI API key](https://arka-ai.ru/docs/account/api-keys). The secret is shown **once** at creation.
## Quick start
```bash
curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/install-arka-provider.sh | bash
```
The script prompts for `ARKA_API_KEY` (blank is allowed; set it later).
Key and default model in one go:
```bash
curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/install-arka-provider.sh \
| bash -s -- --key "$ARKA_API_KEY" --set-default DeepSeek-V4-Pro
```
Then:
1. Open a **new shell** so `export ARKA_API_KEY` is picked up.
2. Restart OpenCode.
3. Run `/models` — you should see provider **Arka-AI** and the chat models.
If OpenCode is started from a GUI, dock icon, or socket rather than a login shell, `~/.bashrc` may not reach it. Set `ARKA_API_KEY` in that process environment, or type it at the installer prompt.
### From inside OpenCode
OpenCode has no plugin store. You do not have to leave the CLI: ask the **current** model (any provider, not Arka) to run the installer via bash.
Paste this (put your real key in):
```
Install the Arka AI OpenCode provider. Do not prompt for the key.
Run this exact command and show the output:
curl -fsSL https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/install-arka-provider.sh | bash -s -- --key 'ARKA_API_KEY_HERE' --set-default DeepSeek-V4-Pro
When it succeeds, tell me I must fully restart OpenCode (config is not hot-reloaded).
```
What happens:
1. The agent uses the built-in bash/shell tool.
2. OpenCode asks permission — Allow the command.
3. The script writes the plugin, skill, and key under `~/.config/opencode` (the installer itself is offline; `curl` only fetches the script).
4. `--key` is required: the interactive key prompt hangs in a non-interactive pipe.
5. **Fully restart OpenCode** — plugins load only at startup. Then `/models` → **Arka-AI**.
Until you restart, this session keeps the old model. That is expected: the installer does not need Arka.
### What the installer writes
| Path | Role |
|------|------|
| `~/.config/opencode/plugins/arka-ai.ts` | **Arka-AI** provider plugin |
| `~/.config/opencode/skills/arka-provider/SKILL.md` | How to refresh the catalog and change the default model |
| `~/.bashrc` (else first existing `.zshrc` / `.profile`) | `export ARKA_API_KEY=…` when a key is given |
| `~/.config/opencode/opencode.json` | Only with `--set-default`: `"model": "Arka-AI/<id>"`. An existing config is **not** replaced. |
The installer is **offline** and idempotent: no network calls; re-running overwrites plugin and skill with the canonical copies.
`python3` is required **only** with `--set-default`. You do not need `bun` on the host — OpenCode ships its own runtime.
### Default model
Use catalog ids with the provider prefix:
```json
{
"model": "Arka-AI/DeepSeek-V4-Pro"
}
```
Examples: `Arka-AI/DeepSeek-V4-Pro`, `Arka-AI/qwen3.8-max`, `Arka-AI/kimi-k3`, `Arka-AI/MiniMax-M3`, `Arka-AI/GLM-5.3`, `Arka-AI/deepseek-v4-flash-0731`.
Those ids stay visible even if the chat filter would drop them.
### Refresh the catalog
```bash
rm -f ~/.cache/opencode/arka-ai-models.json
```
Restart OpenCode. The plugin fetches the public snapshot
`https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/catalog.json`
(no key), then falls back to live `https://api.arka-ai.ru/v1/models`. Local cache TTL is 24 hours. OpenCode does not hot-reload config.
Inspect the published list (no key):
```bash
curl -sS https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/catalog.json | python3 -m json.tool
```
Inference still uses `ARKA_API_KEY` against `https://api.arka-ai.ru/v1`.
### Troubleshooting
| Symptom | Check |
|---------|--------|
| No Arka-AI after `/models` | `~/.config/opencode/plugins/arka-ai.ts` exists; `ARKA_API_KEY` is set **in the OpenCode process**; you opened a new shell |
| Models look stale | `rm -f ~/.cache/opencode/arka-ai-models.json` and restart |
| Plugin import error | Re-run the installer to restore the canonical file |
| `curl` returns HTML | Use `/raw/branch/main/…`, not the repo page |
---
## Files in this folder
| File | Role |
|------|------|
| `install-arka-provider.sh` | Self-contained installer (the one-liner) |
| `plugins/arka-ai.ts` | Plugin source (embedded in the installer) |
| `skills/arka-provider/SKILL.md` | Skill source (embedded in the installer) |
| `catalog.json` | Filtered chat catalog, refreshed daily at 05:00 Europe/Moscow |
| `README.md` | This page — copy into arka-ai.ru «агентные помощники» / agentic helpers |
This directory is `opencode/` inside the public dump [arka/public](https://git.arka-ai.ru/arka/public).
File renamed without changes.
+304
View File
@@ -0,0 +1,304 @@
#!/usr/bin/env bash
#
# install-arka-provider.sh — one-shot installer for the arka-ai.ru LLM provider
# in a fresh opencode install.
#
# It is fully self-contained: it embeds the provider plugin and the companion
# skill, writes them into ~/.config/opencode, and persists ARKA_API_KEY.
#
# Usage:
# bash install-arka-provider.sh # prompt for API key
# ARKA_API_KEY=xxx bash install-arka-provider.sh # key from env (no prompt)
# bash install-arka-provider.sh --key xxx --set-default DeepSeek-V4-Pro
#
# Options:
# --key KEY ARKA_API_KEY to persist (default: $ARKA_API_KEY, else prompt)
# --set-default MODEL also set "model": "Arka-AI/<MODEL>" in opencode.json
# -h, --help show this help
#
# Idempotent: safe to run multiple times. Overwrites plugin/skill with the
# embedded versions; never clobbers an existing opencode.json (only upserts the
# requested default model). Makes no network calls.
set -euo pipefail
usage() {
sed -n '2,21p' "$0"
}
KEY=""
MODEL=""
while [ $# -gt 0 ]; do
case "$1" in
--key) KEY="$2"; shift 2 ;;
--key=*) KEY="${1#*=}"; shift ;;
--set-default) MODEL="$2"; shift 2 ;;
--set-default=*) MODEL="${1#*=}"; shift ;;
-h|--help) usage; exit 0 ;;
*) printf 'unknown option: %s\n' "$1" >&2; usage; exit 1 ;;
esac
done
CONF_DIR="${XDG_CONFIG_HOME:-$HOME/.config}/opencode"
PLUGIN_DIR="$CONF_DIR/plugins"
SKILL_DIR="$CONF_DIR/skills/arka-provider"
mkdir -p "$PLUGIN_DIR" "$SKILL_DIR" "$HOME"
echo "==> writing provider plugin to $PLUGIN_DIR/arka-ai.ts"
cat > "$PLUGIN_DIR/arka-ai.ts" <<'PLUGIN'
import { homedir } from "node:os"
import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs"
import { join } from "node:path"
const PROVIDER_ID = "Arka-AI"
const API_URL = "https://api.arka-ai.ru/v1"
const CATALOG_URL =
"https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/catalog.json"
const CACHE_FILE = join(homedir(), ".cache", "opencode", "arka-ai-models.json")
const TTL_MS = 24 * 60 * 60 * 1000
// Ids that must always be exposed, even if the filter would drop them.
const FORCED_MODELS = [
"DeepSeek-V4-Pro",
"deepseek-v4-flash-0731",
"qwen3.8-max",
"kimi-k3",
"MiniMax-M3",
"GLM-5.3",
]
// Heuristic: drop non-chat models (TTS/ASR/image/video/OCR/embeddings/rerankers…)
// from the catalog so they don't clutter the /models picker.
// Keep in sync with scripts/sync-catalog.py
const NON_CHAT = new RegExp(
[
"tts", "asr", "audio", "speech", "voice", "whisper", "sensevoice",
"image", "(^|[.-])img", "wan", "vidu", "video", "sora", "flux",
"diffus", "kolors", "cogview", "ocr", "embedding", "(^|[.-])embed",
"rerank", "clip", "moderation", "(^|[.-])filter", "detect", "classif",
"(^|[.-])location", "(^|[.-])search", "comfyui", "meme", "hello",
"(^|[.-])sam", "florence", "esrgan", "rmbg", "background", "poster",
"(^|[.-])face", "(^|[.-])lip", "translate", "(^|[.-])mt([2-]|$)",
"(^|[.-])bge", "(^|[.-])e5", "(^|[.-])jina", "(^|[.-])mpnet",
"spark-tts", "cosyvoice", "chattts", "indextts", "step-audio",
"fish-speech", "cosy", "mate-gguf", "tokenizer",
].join("|"),
"i",
)
type ModelMap = Record<string, { name: string }>
type Cache = { ts: number; models: ModelMap }
function loadCache(): Cache | null {
try {
if (existsSync(CACHE_FILE)) return JSON.parse(readFileSync(CACHE_FILE, "utf8"))
} catch {}
return null
}
function saveCache(models: ModelMap): void {
try {
mkdirSync(join(homedir(), ".cache", "opencode"), { recursive: true })
writeFileSync(CACHE_FILE, JSON.stringify({ ts: Date.now(), models }))
} catch {}
}
async function fetchJson(
url: string,
headers: Record<string, string>,
timeoutMs: number,
): Promise<unknown> {
const ctrl = new AbortController()
const timer = setTimeout(() => ctrl.abort(), timeoutMs)
try {
const res = await fetch(url, { headers, signal: ctrl.signal })
if (!res.ok) throw new Error(`HTTP ${res.status}`)
return await res.json()
} finally {
clearTimeout(timer)
}
}
function idsFromPayload(json: unknown): string[] {
const data = (json as { data?: Array<{ id: string }> }).data
return (data ?? []).map((m) => m.id).filter(Boolean)
}
function toModelMap(ids: string[], filter: boolean): ModelMap {
const models: ModelMap = {}
for (const id of ids) {
if (!filter || FORCED_MODELS.includes(id) || !NON_CHAT.test(id)) {
models[id] = { name: id }
}
}
return models
}
async function resolveModels(): Promise<ModelMap | null> {
const cache = loadCache()
if (cache && Date.now() - cache.ts < TTL_MS) return cache.models
try {
const models = toModelMap(idsFromPayload(await fetchJson(CATALOG_URL, {}, 5000)), false)
if (Object.keys(models).length) {
saveCache(models)
return models
}
} catch {}
try {
const key = process.env.ARKA_API_KEY
const json = await fetchJson(
`${API_URL}/models`,
key ? { Authorization: `Bearer ${key}` } : {},
5000,
)
const models = toModelMap(idsFromPayload(json), true)
saveCache(models)
return models
} catch {
return cache?.models ?? null
}
}
export default async function ArkaAIProviderPlugin() {
const models = await resolveModels()
return {
config: (cfg: any) => {
cfg.provider ??= {}
const prev = cfg.provider[PROVIDER_ID] ?? {}
// Static models from opencode.json act as a fallback; the fetched
// catalog is layered on top so the configured default model never
// disappears even if the catalog drops it.
cfg.provider[PROVIDER_ID] = {
name: "Arka-AI",
api: API_URL,
env: ["ARKA_API_KEY"],
...prev,
models: { ...(prev.models ?? {}), ...(models ?? {}) },
}
},
}
}
PLUGIN
echo "==> writing companion skill to $SKILL_DIR/SKILL.md"
cat > "$SKILL_DIR/SKILL.md" <<'SKILL'
---
name: arka-provider
description: Manage the custom Arka-AI (arka-ai.ru) LLM provider in opencode config. Use when the user needs to add, refresh, or fix the arka-ai.ru provider, sync its model catalog, change the Arka-AI default model, or when arka models are missing from /models.
---
# Arka-AI provider
arka-ai.ru is **not** in models.dev, so opencode ships no built-in config for it.
It is wired up locally with a plugin + a static `provider` block in
`~/.config/opencode/opencode.json`.
## How it works
- Plugin: `~/.config/opencode/plugins/arka-ai.ts`
- On startup it loads the published chat catalog
`https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/catalog.json`
(no API key; refreshed daily at 05:00 Europe/Moscow).
- If that fetch fails, it falls back to `https://api.arka-ai.ru/v1/models`
and filters out non-chat models (TTS/ASR/image/video/OCR/embedding/reranker/…)
via a heuristic regex so the `/models` picker stays clean.
- Caches the result at `~/.cache/opencode/arka-ai-models.json` (TTL 24h).
- Injects the provider into config via the `config` hook, layered over the
static block in `opencode.json`.
- Static block (`opencode.json` → `provider["Arka-AI"]`) holds
`name`, `api`, `env`, and a small fallback `models` map that is used when the
catalog is unreachable.
- Inference still uses `ARKA_API_KEY` against `https://api.arka-ai.ru/v1`.
## Refresh the model list
Catalog ~216 raw models / ~116 chat; the public snapshot updates at 05:00
Europe/Moscow. Changes land in opencode only after cache expiry + restart
(config is not hot-reloaded).
1. `rm -f ~/.cache/opencode/arka-ai-models.json`
2. Restart opencode (or run `opencode` again) — the plugin re-fetches on the
next cold start.
Optionally inspect the live API (no key required for listing):
```bash
curl -sS https://api.arka-ai.ru/v1/models | python3 -m json.tool
curl -sS https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/catalog.json | python3 -m json.tool
```
## Change default model
Edit `opencode.json` → `"model": "Arka-AI/<model-id>"`. Use the IDs exactly as
returned by the catalog (e.g. `Arka-AI/DeepSeek-V4-Pro`,
`Arka-AI/qwen3.8-max`, `Arka-AI/kimi-k3`).
## Adjust filtering
Edit `NON_CHAT` (regex) or `FORCED_MODELS` in the plugin file **and** the same
lists in `scripts/sync-catalog.py` in `arka/public`, then re-run the installer
and restart. `FORCED_MODELS` keeps must-have chat models even if the heuristic
would drop them.
SKILL
# --- API key -----------------------------------------------------------
if [ -z "$KEY" ]; then
read -rsp "ARKA_API_KEY (leave blank to skip): " KEY || true
echo
fi
if [ -n "$KEY" ]; then
RC=""
for f in "$HOME/.bashrc" "$HOME/.zshrc" "$HOME/.profile"; do
if [ -f "$f" ]; then RC="$f"; break; fi
done
RC="${RC:-$HOME/.bashrc}"
touch "$RC"
if grep -q 'ARKA_API_KEY' "$RC"; then
echo "==> ARKA_API_KEY already present in $RC (skipped)"
else
printf '\n# opencode Arka-AI provider\nexport ARKA_API_KEY="%s"\n' "$KEY" >> "$RC"
echo "==> wrote ARKA_API_KEY to $RC"
fi
export ARKA_API_KEY="$KEY"
else
echo "==> warning: no ARKA_API_KEY set; export it in your shell before starting opencode" >&2
fi
# --- default model (optional) ------------------------------------------
if [ -n "$MODEL" ]; then
CONF="$CONF_DIR/opencode.json"
echo "==> setting default model Arka-AI/$MODEL in $CONF"
python3 - "$CONF" "Arka-AI/$MODEL" <<'PY'
import json, os, sys
path, model = sys.argv[1], sys.argv[2]
data = {}
if os.path.exists(path):
with open(path) as f:
data = json.load(f)
data.setdefault("$schema", "https://opencode.ai/config.json")
data["model"] = model
with open(path, "w") as f:
json.dump(data, f, indent=2)
f.write("\n")
PY
fi
echo
echo "Done."
echo " - provider plugin: $PLUGIN_DIR/arka-ai.ts"
echo " - skill: $SKILL_DIR/SKILL.md"
echo
echo "Next steps:"
echo " 1. Restart opencode (open a new shell so ARKA_API_KEY is exported)."
echo " 2. Run /models — the Arka-AI catalog (~116 chat models) loads on first start."
echo " 3. Optional: refresh catalog later with: rm -f ~/.cache/opencode/arka-ai-models.json"
@@ -5,7 +5,7 @@ import { join } from "node:path"
const PROVIDER_ID = "Arka-AI" const PROVIDER_ID = "Arka-AI"
const API_URL = "https://api.arka-ai.ru/v1" const API_URL = "https://api.arka-ai.ru/v1"
const CATALOG_URL = const CATALOG_URL =
"https://git.arka-ai.ru/arka/public/raw/branch/main/catalog.json" "https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/catalog.json"
const CACHE_FILE = join(homedir(), ".cache", "opencode", "arka-ai-models.json") const CACHE_FILE = join(homedir(), ".cache", "opencode", "arka-ai-models.json")
const TTL_MS = 24 * 60 * 60 * 1000 const TTL_MS = 24 * 60 * 60 * 1000
@@ -1,9 +1,11 @@
#!/usr/bin/env bash #!/usr/bin/env bash
# Daily catalog refresh for arka/public. Runs on core at 05:00 Europe/Moscow. # Daily catalog refresh for arka/public/opencode.
# Runs on core at 05:00 Europe/Moscow.
set -euo pipefail set -euo pipefail
ROOT="$(cd "$(dirname "$0")/.." && pwd)" # this file lives at opencode/scripts/cron-sync.sh → repo root is ../..
cd "$ROOT" REPO="$(cd "$(dirname "$0")/../.." && pwd)"
cd "$REPO"
export GIT_SSH_COMMAND="${GIT_SSH_COMMAND:-ssh -i /root/.ssh/arka_public_catalog -o IdentitiesOnly=yes -o StrictHostKeyChecking=accept-new}" export GIT_SSH_COMMAND="${GIT_SSH_COMMAND:-ssh -i /root/.ssh/arka_public_catalog -o IdentitiesOnly=yes -o StrictHostKeyChecking=accept-new}"
@@ -11,14 +13,14 @@ git fetch origin main
git checkout -q main git checkout -q main
git reset --hard origin/main git reset --hard origin/main
python3 scripts/sync-catalog.py python3 opencode/scripts/sync-catalog.py
if git diff --quiet -- catalog.json; then if git diff --quiet -- opencode/catalog.json; then
printf '%s no catalog change\n' "$(date -u +%Y-%m-%dT%H:%M:%SZ)" printf '%s no catalog change\n' "$(date -u +%Y-%m-%dT%H:%M:%SZ)"
exit 0 exit 0
fi fi
git add catalog.json git add opencode/catalog.json
git -c user.name='arka-catalog-bot' -c user.email='noreply@arka-ai.ru' \ git -c user.name='arka-catalog-bot' -c user.email='noreply@arka-ai.ru' \
commit -m "chore(catalog): refresh $(date -u +%Y-%m-%d)" commit -m "chore(catalog): refresh $(date -u +%Y-%m-%d)"
git push origin main git push origin main
@@ -1,7 +1,7 @@
#!/usr/bin/env python3 #!/usr/bin/env python3
"""Rebuild catalog.json from the public arka-ai.ru OpenAI-compatible models list. """Rebuild catalog.json from the public arka-ai.ru OpenAI-compatible models list.
Keep FORCED_MODELS / NON_CHAT in lockstep with plugins/arka-ai.ts. Keep FORCED_MODELS / NON_CHAT in lockstep with opencode/plugins/arka-ai.ts.
""" """
from __future__ import annotations from __future__ import annotations
@@ -13,7 +13,7 @@ It is wired up locally with a plugin + a static `provider` block in
- Plugin: `~/.config/opencode/plugins/arka-ai.ts` - Plugin: `~/.config/opencode/plugins/arka-ai.ts`
- On startup it loads the published chat catalog - On startup it loads the published chat catalog
`https://git.arka-ai.ru/arka/public/raw/branch/main/catalog.json` `https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/catalog.json`
(no API key; refreshed daily at 05:00 Europe/Moscow). (no API key; refreshed daily at 05:00 Europe/Moscow).
- If that fetch fails, it falls back to `https://api.arka-ai.ru/v1/models` - If that fetch fails, it falls back to `https://api.arka-ai.ru/v1/models`
and filters out non-chat models (TTS/ASR/image/video/OCR/embedding/reranker/…) and filters out non-chat models (TTS/ASR/image/video/OCR/embedding/reranker/…)
@@ -40,7 +40,7 @@ Optionally inspect the live API (no key required for listing):
```bash ```bash
curl -sS https://api.arka-ai.ru/v1/models | python3 -m json.tool curl -sS https://api.arka-ai.ru/v1/models | python3 -m json.tool
curl -sS https://git.arka-ai.ru/arka/public/raw/branch/main/catalog.json | python3 -m json.tool curl -sS https://git.arka-ai.ru/arka/public/raw/branch/main/opencode/catalog.json | python3 -m json.tool
``` ```
## Change default model ## Change default model