AIsa Provider for OpenClaw

View on GitHub -> Use AIsa as an OpenClaw model provider. Configure OpenClaw to access Qwen, DeepSeek, Kimi, GLM, and other models through one gateway.

Install

First, install the AIsa CLI if you have not already:

npm install -g @aisa-one/cli

Then install the skill:

aisa skills install aisa-provider

What can agents do with it?

OpenClaw setup

Configure OpenClaw with the AIsa provider.

Chinese LLM routing

Use Qwen, DeepSeek, GLM, and other Chinese models.

Global models

Route prompts to GPT, Claude, Gemini, and more.

Agent runtime defaults

Standardize model selection for OpenClaw workflows.

Quick Setup

Option 1: Environment Variable (fastest)

export AISA_API_KEY="your-key-here"

OpenClaw auto-detects AISA_API_KEY and registers AIsa as a provider. No config file changes needed.

Option 2: Interactive Onboarding

openclaw onboard --auth-choice aisa-api-key

Option 3: CLI with Key

openclaw onboard --auth-choice aisa-api-key --aisa-api-key "your-key-here"

Option 4: Manual Config in `~/.openclaw/openclaw.json`

{
  "models": {
    "providers": {
      "aisa": {
        "baseUrl": "https://api.aisa.one/v1",
        "apiKey": "${AISA_API_KEY}",
        "api": "openai-completions",
        "models": [
          {
            "id": "aisa/qwen3-max",
            "name": "Qwen3 Max",
            "reasoning": true,
            "input": ["text", "image"],
            "contextWindow": 262144,
            "maxTokens": 16384,
            "supportsDeveloperRole": false,
            "cost": {
              "input": 0.72,
              "output": 3.60,
              "cacheRead": 0.72,
              "cacheWrite": 0
            }
          },
          {
            "id": "aisa/qwen-plus-2025-12-01",
            "name": "Qwen Plus",
            "reasoning": true,
            "input": ["text", "image"],
            "contextWindow": 1000000,
            "maxTokens": 16384,
            "supportsDeveloperRole": false,
            "cost": {
              "input": 0.28,
              "output": 0.84,
              "cacheRead": 0.28,
              "cacheWrite": 0
            }
          },
          {
            "id": "aisa/qwen-mt-flash",
            "name": "Qwen MT Flash",
            "reasoning": true,
            "input": ["text"],
            "contextWindow": 1000000,
            "maxTokens": 8192,
            "supportsDeveloperRole": false,
            "cost": {
              "input": 0.072,
              "output": 0.2205,
              "cacheRead": 0.072,
              "cacheWrite": 0
            }
          },
          {
            "id": "aisa/deepseek-v3.1",
            "name": "DeepSeek V3.1",
            "reasoning": true,
            "input": ["text"],
            "contextWindow": 262144,
            "maxTokens": 8192,
            "supportsDeveloperRole": false,
            "cost": {
              "input": 0.4018,
              "output": 1.2047,
              "cacheRead": 0.4018,
              "cacheWrite": 0
            }
          },
          {
            "id": "aisa/kimi-k2.5",
            "name": "Kimi K2.5",
            "reasoning": true,
            "input": ["text"],
            "contextWindow": 262144,
            "maxTokens": 8192,
            "supportsDeveloperRole": false,
            "cost": {
              "input": 0.4018,
              "output": 2.1077,
              "cacheRead": 0.4018,
              "cacheWrite": 0
            }
          }
        ]
      }
    }
  },
  "agents": {
    "defaults": {
      "model": {
        "primary": "aisa/qwen3-max"
      }
    }
  }
}

Available Models

Default Models (pre-configured, API-verified ✅)

Model	Model ID	Best For	Context	Reasoning	Verified
Qwen3 Max	`aisa/qwen3-max`	Complex reasoning, flagship tasks	256K	✅	✅
Qwen Plus	`aisa/qwen-plus-2025-12-01`	Main production model	1M	✅	✅
Qwen MT Flash	`aisa/qwen-mt-flash`	Translation-oriented lightweight tasks	1M	✅	✅
DeepSeek V3.1	`aisa/deepseek-v3.1`	Cost-effective reasoning	256K	✅	✅
Kimi K2.5	`aisa/kimi-k2.5`	Vision, visual coding, long-video understanding	256K	✅	✅

Moonshot’s newest generation, aisa/kimi-k3, and its coding-oriented aisa/kimi-k2.7-code route are also available — they are not pre-configured above, so add them to your config manually if you want them.

⭐ Kimi K2.5 — Enterprise ZDR route

Kimi K2.5 is the Moonshot route covered by AIsa’s enterprise Zero Data Retention agreement, priced below official Moonshot rates. Key features:

Strong reasoning and multi-step problem solving
Competitive performance on coding and math benchmarks
Available through AIsa’s Moonshot partnership

🔒 Zero Data Retention (ZDR) — Enterprise-Grade Privacy

Accessing Kimi K2.5 through AIsa comes with enterprise-level Zero Data Retention guarantees. Users do not need to worry about data privacy — AIsa has executed a formal ZDR agreement with Moonshot AI. Under the Supplemental Enterprise Service Agreement (dated 10 February 2026) between AIsa and Kimi (Moonshot AI PTE. LTD.):

Customer data is NOT retained by Moonshot after processing
Generated outputs are NOT stored on Moonshot’s infrastructure
No data is used for model training — your prompts and completions remain private
Processing is governed by contractual enterprise terms, not consumer ToS

This makes AIsa the recommended path for privacy-sensitive or enterprise workloads that require access to Kimi K2.5. When calling Kimi K2.5 directly via Moonshot’s consumer API, standard consumer data policies apply — by routing through AIsa, your data benefits from the negotiated ZDR protections. ⚠️ Important: Temperature restriction Kimi K2.5 only accepts temperature=1.0. Using any other value will return an error:

Error: invalid temperature: only 1 is allowed for this model

If your OpenClaw config or agent sets a different temperature, override it for Kimi:

/model aisa/kimi-k2.5

OpenClaw will use the model’s default temperature when not explicitly set. Kimi K2.5 pricing on AIsa (per 1M tokens): $0.4018 input / $2.1077 output, cache read $0.4018.

Compare against Moonshot’s current official rates on their own pricing page — AIsa prices this route below retail, but the exact gap moves whenever either side changes. Live AIsa rates: https://console.aisa.one/pricing

Additional Models Available via AIsa

Users can add any model supported by AIsa to their config. The gateway serves 108 models as of August 5, 2026: Qwen family (24 models, including Wan image/video):

qwen3.8-max, qwen3.7-max, qwen3.7-max-2026-06-08, qwen3.7-plus, qwen3.7-flash, qwen3.7-flash-2026-07-15, qwen3-max, qwen-plus-2025-12-01, qwen-flash
qwen3-coder-plus, qwen3-coder-flash, qwen3-coder-480b-a35b-instruct
qwen3.6-plus, qwen3.6-plus-2026-04-02
qwen3-vl-plus, qwen3-vl-flash, qwen3-vl-flash-2025-10-15 (vision models)
qwen-mt-flash, qwen-mt-lite (translation models)
wan2.7-image, wan2.7-image-pro, wan2.7-t2v, wan2.7-i2v, wan2.7-r2v (media models)

DeepSeek (7 models):

deepseek-v4-pro, deepseek-v4-flash, deepseek-v4-flash-0731, deepseek-v3.2, deepseek-v3.1, deepseek-v3, deepseek-r1

Kimi / Moonshot (5 models):

kimi-k3, kimi-k2.7-code, kimi-k2.6, kimi-k2.5, kimi-k2-thinking

Also available: GPT series (27), Claude series (12), xAI Grok (5), Zhipu GLM (4), ByteDance Seed/Seedream/Dreamina (10), StepFun (3), MiniMax (2), Gemini (2), Xiaomi MiMo (2), and more. List all available models:

curl https://api.aisa.one/v1/models -H "Authorization: Bearer $AISA_API_KEY"

Model ID Versioning

AIsa uses versioned model IDs for some models. If you encounter a 503 - No available channels error, the model ID may need updating. Known model ID mappings:

Common Name	Correct AIsa Model ID	❌ Does NOT work
Qwen Plus	`qwen-plus-2025-12-01`	`qwen3-plus`, `qwen-plus`, `qwen-plus-latest`
Qwen Flash	`qwen-flash`	`qwen3-flash`, `qwen-turbo`, `qwen-turbo-latest`
Qwen MT Flash (translation)	`qwen-mt-flash`	`qwen-mt`, `qwen-translate`
Qwen Max	`qwen3-max`	(works as-is)
DeepSeek V3.1	`deepseek-v3.1`	(works as-is)
Kimi K2.5	`kimi-k2.5`	(works as-is)
Kimi K3	`kimi-k3`	`kimi-k3-latest`, `kimi-3`

To check the latest available model IDs:

curl https://api.aisa.one/v1/models -H "Authorization: Bearer $AISA_API_KEY"

Switching Models

In chat (TUI):

/model aisa/qwen3-max
/model aisa/deepseek-v3.1
/model aisa/kimi-k2.5

Via CLI:

openclaw models set aisa/qwen3-max

AIsa pricing for the default models (per 1M tokens)

Public AIsa rates as of July 28, 2026. Real-time pricing is subject to change — always check https://console.aisa.one/pricing for the latest rates, and each vendor’s own page for their retail rates.

Model	Input	Output	Cache read
`aisa/qwen-mt-flash`	$0.0720	$0.2205	$0.0720
`aisa/qwen-plus-2025-12-01`	$0.2800	$0.8400	$0.2800
`aisa/qwen3-max`	$0.7200	$3.6000	$0.7200
`aisa/deepseek-v3.1`	$0.4018	$1.2047	$0.4018
`aisa/kimi-k2.5`	$0.4018	$2.1077	$0.4018
`aisa/kimi-k3`	$3.0000	$15.0000	$0.3000

For many non-Anthropic models AIsa negotiates volume discounts with the upstream provider and passes the savings through — see AI Model Pricing for the full catalog.

Official Partnerships

AIsa maintains verified partnerships with:

Alibaba Cloud — Qwen Key Account (full model family, 3 global regions: CN, US-Virginia, Singapore)
BytePlus — Doubao by ByteDance
DeepSeek — via Alibaba Cloud integration
Moonshot — Kimi integration (kimi-k3, kimi-k2.7-code, kimi-k2.6, kimi-k2.5, kimi-k2-thinking), with an enterprise Zero Data Retention (ZDR) agreement covering Kimi K2.5 (effective Feb 10, 2026)

Qwen Region Support

AIsa provides access to Qwen models across 3 global regions via Alibaba Cloud:

🇨🇳 China (default)
🇺🇸 US (Virginia)
🇸🇬 Singapore

This is unique to AIsa’s Key Account status. Other providers like OpenRouter or the free Qwen Portal typically route through CN only.

Response Latency (tested Feb 2026)

Model	Avg Latency	Rating
Qwen3 Max	~1,577 ms	⭐⭐⭐⭐⭐ Fastest
Qwen MT Flash	~1,918 ms	⭐⭐⭐⭐ Fast
Kimi K2.5	~2,647 ms	⭐⭐⭐ Medium
DeepSeek V3.1	~3,002 ms	⭐⭐⭐ Medium
Qwen Plus	~8,207 ms	⭐⭐ Slower

Troubleshooting

”503 - No available channels” error

The model ID may be incorrect or outdated. Check the Model ID Versioning section above for correct IDs. Common fixes:

qwen3-plus -> use qwen-plus-2025-12-01
qwen3-flash -> use qwen-flash (or qwen-mt-flash for translation)

“Model not found” error

Ensure the model ID uses the aisa/ prefix in OpenClaw config:

✅ aisa/qwen3-max
❌ qwen3-max

Kimi K2.5 “invalid temperature” error

Kimi K2.5 only accepts temperature=1.0. If your config sets a different temperature, add a model-specific override or let OpenClaw use the default.

Kimi K2.5 empty response

In rare cases Kimi K2.5 may return empty content while consuming output tokens. Retry the request — this is typically transient.

API key not detected

Check env var: echo $AISA_API_KEY
Or verify in config: openclaw config get auth.profiles
Re-run onboarding: openclaw onboard --auth-choice aisa-api-key

Streaming not working

AIsa uses the OpenAI-compatible API (openai-completions). Ensure your config has:

"api": "openai-completions"

Rate limits or daily caps

AIsa has no daily request limits (unlike the free Qwen Portal which caps at 2,000 req/day).

Get an API Key

Visit https://console.aisa.one/
Sign up and create an API key
Set it as AISA_API_KEY or use the onboarding wizard

Notes

AIsa’s endpoint is OpenAI-compatible (https://api.aisa.one/v1)
All models support streaming and function calling
supportsDeveloperRole is set to false for Qwen models
Context windows vary per model; see the model catalog rather than assuming a default
Reasoning (thinking) is enabled for all default models
Kimi K2.5 requires temperature=1.0 — other values cause API errors
Kimi K2.5 via AIsa is covered by enterprise Zero Data Retention (ZDR) — data is not retained or used for training
Image/Video generation models (WAN) are available but require separate configuration
AIsa API serves 108 models as of August 5, 2026 — use the models endpoint to discover the current list

Get started

Sign up at aisa.one (new accounts start with $2 free credit).
Generate an API key from the console.

Set your key and install the skill:

export AISA_API_KEY="your-key"
npm install -g @aisa-one/cli
aisa skills install aisa-provider

Start a new agent session so the runtime loads the updated skill instructions.

Quick Setup: How to Use AIsa in OpenClaw

Recommended OpenClaw setup path.

AIsa CN-LLM Route

Route Chinese-language prompts to Chinese LLMs.

AIsa LLM Router

Route prompts across supported models.

​Install

​What can agents do with it?

OpenClaw setup

Chinese LLM routing

Global models

Agent runtime defaults

​Quick Setup

​Option 1: Environment Variable (fastest)

​Option 2: Interactive Onboarding

​Option 3: CLI with Key

​Option 4: Manual Config in ~/.openclaw/openclaw.json

​Available Models

​Default Models (pre-configured, API-verified ✅)

​⭐ Kimi K2.5 — Enterprise ZDR route

​🔒 Zero Data Retention (ZDR) — Enterprise-Grade Privacy

​Additional Models Available via AIsa

​Model ID Versioning

​Switching Models

​AIsa pricing for the default models (per 1M tokens)

​Official Partnerships

​Qwen Region Support

​Response Latency (tested Feb 2026)

​Troubleshooting

​”503 - No available channels” error

​“Model not found” error

​Kimi K2.5 “invalid temperature” error

​Kimi K2.5 empty response

​API key not detected

​Streaming not working

​Rate limits or daily caps

​Get an API Key

​Notes

​Get started

​Related