oh-my-coding-maas-gateway

LiteLLM proxy routing Huawei MaaS models to opencode, Codex CLI, Claude
Code CLI, and Pi agent — with virtual keys, multi-key load balancing,
dual-format endpoints, and Prometheus + Grafana observability.

What Is This?

A self-hosted gateway that sits between your coding tools and Huawei ModelArts
MaaS, giving you load balancing, virtual keys, budget tracking, and
observability — all through a single local proxy.

  Tools               LiteLLM (:4000)                  Huawei MaaS
  ─────               ───────────────                  ────────────

  opencode    ──→ /v1/chat/completions ──→ openai/ provider    ──→ MaaS OpenAI endpoint
  Codex CLI   ──→ /v1/responses        ──→ openai/ provider    ──→ MaaS OpenAI endpoint
  Claude Code ──→ /v1/messages         ──→ anthropic/ provider ──→ MaaS Anthropic endpoint
  Pi agent    ──→ /v1/chat/completions ──→ openai/ provider    ──→ MaaS OpenAI endpoint

  LiteLLM: load-balances across N MaaS keys · PostgreSQL (:5432)
  Observability: LiteLLM ──/metrics──→ Prometheus (:9090) ──→ Grafana (:3000)

4 models: glm-5.3, glm-5.2, glm-5.1, deepseek-v4.1-flash-1017 (accepts image
input — first multimodal model)

Install

The bootstrap script is the only installation method. It handles
everything — prerequisites, Docker stack, coding tools, virtual keys,
validation, and optional companion skill.

curl -fsSL https://raw.githubusercontent.com/wallacelw/oh-my-coding-maas-gateway/main/scripts/bootstrap.sh | bash

Bootstrap will prompt for your Huawei MaaS API key, show an interactive
menu to select which coding tools to install, auto-install prerequisites,
deploy the proxy + observability stack, configure each tool with its own
virtual key, run validation, and offer to install the companion skill.

Non-interactive

# Fresh install with API key (no prompts)
curl -fsSL https://raw.githubusercontent.com/wallacelw/oh-my-coding-maas-gateway/main/scripts/bootstrap.sh | bash -s -- -y --api-key=sk-xxxx

# Upgrade existing install (pull updates, regenerate configs)
curl -fsSL https://raw.githubusercontent.com/wallacelw/oh-my-coding-maas-gateway/main/scripts/bootstrap.sh | bash -s -- -y

# Non-interactive with specific tools only
curl -fsSL https://raw.githubusercontent.com/wallacelw/oh-my-coding-maas-gateway/main/scripts/bootstrap.sh | bash -s -- -y --api-key=sk-xxxx --tool=litellm,opencode

Non-interactive upgrades discard local repo changes if git pull fails —
bootstrap prints a warning counting what is being discarded before
resetting to origin/main. .env, the generated LiteLLM config, and
Docker volumes are preserved; tracked repo files, including config
templates, are reset.

Estimated time: ~5 min fresh, ~2 min upgrade. For flags and non-interactive
usage, see INSTALLATION.md.

After install

opencode          # or:  codex  or:  claude --bare  or:  pi

Monitor at http://localhost:3000 (Grafana) or http://localhost:4000/ui
(LiteLLM Admin).

Upgrade

Same command — bootstrap detects existing install, compares versions, and
pulls updates. All secrets and data preserved.

curl -fsSL https://raw.githubusercontent.com/wallacelw/oh-my-coding-maas-gateway/main/scripts/bootstrap.sh | bash

Or non-interactive:

curl -fsSL https://raw.githubusercontent.com/wallacelw/oh-my-coding-maas-gateway/main/scripts/bootstrap.sh | bash -s -- -y

In non-interactive mode, if git pull fails, bootstrap warns about the
local commits and uncommitted changes it is discarding, then resets to
origin/main — .env, the generated LiteLLM config, and Docker volumes
are preserved; tracked repo files, including config templates, are reset.

After upgrade, restart any running coding tools — plugin/preset changes
are not hot-reloaded.

Update coding tools only

To check and update individual components (opencode, Codex, Claude Code,
Pi, LiteLLM, Grafana, Prometheus; PostgreSQL is listed as pinned and
display-only, never auto-updated) without re-running the full pipeline:

./scripts/update.sh              # interactive: show versions, select which to update
./scripts/update.sh --check      # show version table only
./scripts/update.sh --all        # update all with updates available
./scripts/update.sh --dry-run    # show what would be updated, no changes

Does not touch passwords, API keys, or virtual keys. Note: updating
Docker images or slim plugin modifies docker-compose.yml and
scripts/03a_opencode.sh in-place — changes are auto-committed.

Companion Skill

After installation, bootstrap offers to install
SKILL.md as a skill into each detected coding agent. This
gives your agents operational guidance — health diagnosis, key/model
management, debug routing, observability, and recovery.

./scripts/05_skill.sh --yes    # install manually anytime

Uninstall

./scripts/uninstall.sh --all --dry-run   # preview
./scripts/uninstall.sh --tool=opencode   # remove one agent
./scripts/uninstall.sh --all             # remove everything

Backup

./scripts/06_backup.sh                  # dump LiteLLM DB (spend history, keys, budgets)
./scripts/06_backup.sh --restore FILE   # restore from a dump
./scripts/07_dashboard_shots.sh         # capture Grafana dashboard screenshots (visual verification)

Run before docker compose down -v — the DB volume is destroyed otherwise.

Documentation