NagaAI

OpenClaw provider plugin for NagaAI, an OpenAI-compatible API for models from many labs.

Install

openclaw plugins install clawhub:@nagaai/openclaw-provider

NagaAI provider for OpenClaw

Adds NagaAI to OpenClaw as the nagaai provider: models from OpenAI, Anthropic, Google, DeepSeek, Moonshot, Z.ai, xAI and others through one API. The model list, context windows and prices come from NagaAI's public catalog, so new models show up without a plugin update.

Property	Value
Provider id	`nagaai`
Package	`@nagaai/openclaw-provider`
API key env var	`NAGA_API_KEY`
Onboarding flag	`--auth-choice nagaai-api-key`
Direct CLI flag	`--nagaai-api-key <key>`
API	OpenAI Responses (Chat Completions optional)
Base URL	`https://api.naga.ac/v1`
Default model	`nagaai/anthropic/claude-sonnet-5.5`

Requirements

OpenClaw 2026.11.2 or later.
A NagaAI API key from the API Keys page.

Install

bash

openclaw plugins install clawhub:@nagaai/openclaw-provider

Set the API key

Run onboarding and pick NagaAI:

bash

openclaw onboard --auth-choice nagaai-api-key

Or sign in for this provider only:

bash

openclaw models auth login --provider nagaai --set-default

Or set the environment variable for the Gateway process:

bash

export NAGA_API_KEY="ng-..."

Select a model

bash

openclaw models list --provider nagaai --all --refresh
openclaw models set nagaai/anthropic/claude-sonnet-5.5

Write the NagaAI model id after nagaai/, for example nagaai/openai/gpt-6.1-sol or nagaai/deepseek/deepseek-v4.1-flash. The full catalog is at https://naga.ac/models.

How the model list works

With a key set, the plugin reads GET https://api.naga.ac/v1/models and keeps the models whose chat operation is served on the API the plugin uses and takes tools. The request carries no key. The result is cached for 60 seconds. If the request fails, OpenClaw gets a built-in list of six models.

Each catalog row maps to an OpenClaw model:

OpenClaw field	From `/v1/models`
`name`	`display_name`
`contextWindow`	`operations[chat].limits.context_window`
`input`	`["text", "image"]` when `architecture.input_modalities` has `image`
`reasoning`	`true` when `operations[chat].parameters` has `reasoning_effort`
`cost`	unconditional `pricing` rows, from USD per token to USD per million tokens

NagaAI accepts the reasoning effort scale none to xhigh for every reasoning model and maps each level to the nearest one the model supports.

/v1/models publishes no per-model output limit, so every model gets maxTokens 8192, the value OpenClaw uses for discovered models. To allow longer answers from one model, set maxTokens in its params:

json5

{
  agents: {
    defaults: {
      models: {
        "nagaai/anthropic/claude-sonnet-5.5": { params: { maxTokens: 32000 } },
      },
    },
  },
}

Responses and Chat Completions

The plugin uses the Responses API. NagaAI keeps no conversation state, so OpenClaw sends the full conversation on every turn, as it does for any custom Responses endpoint. To switch every NagaAI model to Chat Completions:

bash

openclaw config set plugins.entries.nagaai.config.api openai-completions

Disclosure

Your prompts and API key go to api.naga.ac. NagaAI is a multi-model API that forwards requests to upstream model providers and bills in prepaid credits.
The plugin makes one network call of its own, the catalog fetch to https://api.naga.ac/v1/models, without your key. It adds no attribution or tracking headers, writes no files and runs no shell commands.

Development

bash

npm ci
npm run check            # type check, tests, build, ClawHub validation
NAGA_LIVE=1 npm test     # also checks the built-in models against the live catalog

Docs: https://docs.naga.ac/integrations/agents/openclaw · License: MIT