OpenRouter Provider

Access 100+ AI models through OpenRouter’s unified API and run observation extraction off-plan

​

OpenRouter Provider

Claude-mem supports OpenRouter as an alternative provider for observation extraction. OpenRouter provides a unified API to access 100+ models from different providers including Google, Meta, Mistral, DeepSeek, and many others.

Memory runs off-plan: With OpenRouter (or the claude-mem observer, which runs on this provider), observation extraction happens off your Claude plan — so memory work never shares your plan usage.

claude-mem observer: When you sign in during npx claude-mem install and choose the claude-mem observer, claude-mem configures this provider for you automatically with a provisioned key — no manual setup needed. The rest of this page covers bringing your own OpenRouter key.

​

Why Use OpenRouter?

• Runs off-plan: Observation extraction never shares your Claude plan usage
• Access to 100+ models: Choose from models across multiple providers through one API
• Free model options: Several high-quality models are available at no charge
• Errors throw clearly: 429s, 5xx, and network failures throw — leaving messages pending so they can be retried
• Hot-swappable: Switch providers without restarting the worker
• Multi-turn conversations: Full conversation history maintained across API calls

​

Free Models on OpenRouter

OpenRouter actively supports democratizing AI access by offering free models. These are production-ready models suitable for observation extraction.

​

Featured Free Models

Model	ID	Parameters	Context	Best For
Cohere North Mini Code	cohere/north-mini-code:free	30B (3B active, MoE)	256K	Default; agentic coding
Poolside Laguna XS 2.1	poolside/laguna-xs-2.1:free	33B (3B active, MoE)	256K	Coding agents
NVIDIA Nemotron 3 Ultra	nvidia/nemotron-3-ultra-550b-a55b:free	550B (55B active, MoE)	1M	Reasoning, orchestration

Default Model: Claude-mem uses cohere/north-mini-code:free by default, a sparse mixture-of-experts coding model with a 256K context window.

​

Free Model Considerations

• Rate limits: Free models may have stricter rate limits than paid models
• Availability: Free capacity depends on provider partnerships and demand
• Retirement: OpenRouter retires free model ids regularly. If the default is retired, claude-mem moves stored settings off it automatically. If a model you picked is retired, the observer reports “OpenRouter model unavailable” (see Troubleshooting)
• Queue times: During peak usage, requests may be queued briefly
• Max tokens: Most free models support 65,536 completion tokens

All free models support:

• Tool use and function calling
• Temperature and sampling controls
• Stop sequences
• Streaming responses

​

Getting an API Key

1. Go to OpenRouter
2. Sign in with Google, GitHub, or email
3. Navigate to API Keys
4. Click Create Key
5. Copy and securely store your API key

Free to start: No credit card required to create an account or use free models. Add credits only if you want to use premium models.

​

Configuration

​

Settings

Setting	Values	Default	Description
CLAUDE_MEM_PROVIDER	claude, codex, gemini, openrouter, openai-compatible	claude	AI provider for observation extraction
CLAUDE_MEM_OPENROUTER_API_KEY	string	—	Your OpenRouter API key
CLAUDE_MEM_OPENROUTER_MODEL	string or array	cohere/north-mini-code:free	Model identifier, or a list to fall back through (see Model fallbacks)
CLAUDE_MEM_OPENROUTER_BASE_URL	string	—	Optional: custom OpenAI-compatible base URL, set in the settings file or environment only (see Custom OpenAI-compatible gateways)
CLAUDE_MEM_OPENROUTER_SITE_URL	string	—	Optional: URL for analytics attribution
CLAUDE_MEM_OPENROUTER_REASONING_EFFORT	none, minimal, low, medium, high	—	Optional: reasoning effort for openrouter.ai models. none turns reasoning off, the fix for a model that spends the output budget thinking and answers with nothing. Every other value is sent as a reasoning token cap (reasoning.max_tokens: 455, 1,024, 4,096 or 16,384 tokens, matching OpenRouter’s effort shares at the default 4,096-token answer cap) and added to max_tokens, so thinking cannot use the answer’s budget. Unset sends nothing. Never sent to custom base URLs or the claude-mem observer
CLAUDE_MEM_OPENROUTER_APP_NAME	string	claude-mem	Optional: App name for analytics
CLAUDE_MEM_OPENROUTER_TEMPERATURE	Number, 0–2	Unset (0.3)	Optional worker sampling temperature, e.g. 0.7 or "0.7". Settings file or environment; environment takes precedence. Overrides extra-body temperature when set. Blank or invalid values preserve existing behavior
CLAUDE_MEM_OPENROUTER_EXTRA_BODY	JSON object	—	Optional: extra request fields (see Extra request fields). Settings file or environment only

​

Using the Settings UI

1. Open the worker URL printed on startup
2. Click the gear icon to open Settings
3. Under AI Provider, select OpenRouter
4. Enter your OpenRouter API key
5. Optionally select a different model

Settings are applied immediately—no restart required.

The base URL is read-only in the viewer. Configure it in ~/.claude-mem/settings.json or CLAUDE_MEM_OPENROUTER_BASE_URL in the worker’s environment. It chooses where your API key and prompt content are sent, so the settings HTTP API cannot change it. Saving other settings preserves it.

​

Manual Configuration

Edit ~/.claude-mem/settings.json:

{
  "CLAUDE_MEM_PROVIDER": "openrouter",
  "CLAUDE_MEM_OPENROUTER_API_KEY": "sk-or-v1-your-key-here",
  "CLAUDE_MEM_OPENROUTER_MODEL": "cohere/north-mini-code:free"
}

​

Model fallbacks

CLAUDE_MEM_OPENROUTER_MODEL also accepts a list of model ids. The first is the one normally used; the rest are handed to OpenRouter’s native models fallback array, which tries them in order when the one before it errors.

{
  "CLAUDE_MEM_PROVIDER": "openrouter",
  "CLAUDE_MEM_OPENROUTER_API_KEY": "sk-or-v1-your-key-here",
  "CLAUDE_MEM_OPENROUTER_MODEL": [
    "vendor/fast-model:free",
    "vendor/backup-model:free",
    "vendor/paid-model"
  ]
}

A comma- or space-separated string works too, so "vendor/a, vendor/b" means the same thing as the two-entry array.

Fallbacks are sent only to openrouter.ai. A custom CLAUDE_MEM_OPENROUTER_BASE_URL gateway speaks plain OpenAI, which has no models field, so it receives the first id and ignores the rest.

Alternatively, set the API key via environment variable:

export OPENROUTER_API_KEY="sk-or-v1-your-key-here"

The settings file takes precedence over the environment variable.

​

Custom OpenAI-compatible gateways

The OpenRouter provider is a generic OpenAI-compatible client. Set CLAUDE_MEM_OPENROUTER_BASE_URL to any server that speaks /chat/completions, such as DeepSeek (https://api.deepseek.com), LM Studio (http://localhost:1234/v1) or IO Intelligence by io.net. The model id is sent as-is, and the first id of a fallback list is used (plain OpenAI has no models field). Use the key that endpoint issued: a claude-mem observer key (cm_pro_…) is only ever sent to the claude-mem observer.

The OpenAI-compatible provider does the same job on settings of its own, with presets for common endpoints, and sends none of OpenRouter’s attribution headers. Prefer it for an endpoint that is not OpenRouter.

For example, IO Intelligence:

{
  "CLAUDE_MEM_PROVIDER": "openrouter",
  "CLAUDE_MEM_OPENROUTER_BASE_URL": "https://api.intelligence.io.solutions/api/v1",
  "CLAUDE_MEM_OPENROUTER_MODEL": "meta-llama/Llama-3.3-70B-Instruct",
  "CLAUDE_MEM_OPENROUTER_API_KEY": "YOUR_IO_INTELLIGENCE_API_KEY"
}

Model ids are org/name pairs from the IO Intelligence model list, for example deepseek-ai/DeepSeek-R1-0528 or zai-org/GLM-4.6. A trailing slash on the base URL is fine; /chat/completions is appended automatically.

​

Extra request fields

CLAUDE_MEM_OPENROUTER_EXTRA_BODY merges a JSON object into every request, for provider-specific fields claude-mem does not set itself. The common case is a reasoning model that spends the whole output budget thinking and answers with nothing. Turn its reasoning off:

{
  "CLAUDE_MEM_OPENROUTER_EXTRA_BODY": { "reasoning": { "enabled": false } }
}

As an environment variable it is the same object as a string: CLAUDE_MEM_OPENROUTER_EXTRA_BODY='{"reasoning":{"enabled":false}}'. A custom gateway may use its vendor’s own field instead, for example {"thinking":{"type":"disabled"}}.

• The value must be a JSON object. Anything else is ignored with a warning in the worker log.
• It cannot replace model, models, messages, stream, max_tokens or max_completion_tokens. Those fields are dropped with a warning. Set the output cap with CLAUDE_MEM_OBSERVER_MAX_OUTPUT_TOKENS.
• It is never sent to the claude-mem observer.
• When CLAUDE_MEM_OPENROUTER_REASONING_EFFORT is set, it decides reasoning and the extra body’s reasoning field is ignored.
• Set it in ~/.claude-mem/settings.json or the environment. The settings UI and API do not write it.

​

Model Selection Guide

​

Free Models

Recommended: cohere/north-mini-code:free (the default)

• 256K context window handles large observations
• 64K max completion tokens
• Sparse mixture-of-experts architecture (3B active parameters)

Alternatives:

• poolside/laguna-xs-2.1:free - Coding-agent model, 256K context
• nvidia/nemotron-3-ultra-550b-a55b:free - 1M context, strong reasoning

​

Premium Models (Higher Quality/Speed)

Model	Best For
anthropic/claude-haiku-4.5	Same model as the Claude provider’s default
google/gemini-2.5-flash	Fast responses
openai/gpt-4o	GPT-4 quality

Premium models draw on your OpenRouter credit — still off-plan, so they never touch your Claude plan usage.

​

Context Window Management

The full conversation history is sent to the model on each request. claude-mem does not apply any client-side truncation or message-count cap — the provider and model own their own context window. If you need to bound context for a specific model, choose a model with an appropriate context length or manage limits at the OpenRouter/provider level.

​

Usage Tracking

Logs include detailed usage information:

OpenRouter API usage: {
  model: "cohere/north-mini-code:free",
  inputTokens: 2500,
  outputTokens: 1200,
  totalTokens: 3700,
  estimatedCostUSD: "0.00",
  messagesInContext: 8
}

​

Provider Switching

You can switch between providers at any time:

• No restart required: Changes take effect on the next observation
• Conversation history preserved: When switching mid-session, the new provider sees the full conversation context
• Seamless transition: All providers use the same observation format

​

Switching via UI

1. Open Settings in the viewer
2. Change the AI Provider dropdown
3. The next observation will use the new provider

​

Switching via Settings File

{
  "CLAUDE_MEM_PROVIDER": "openrouter"
}

​

Error Behavior

If OpenRouter errors, claude-mem logs the failure and re-throws so the message stays pending for later retry. There is no Claude SDK fallback — earlier docs claimed automatic Claude fallback, but the wiring was never actually engaged in production (#2087). To switch providers, change CLAUDE_MEM_PROVIDER in settings.

Throwing conditions:

• Rate limiting (HTTP 429)
• Server errors (HTTP 500, 502, 503)
• Network issues (connection refused, timeout)
• 4xx errors other than 429
• Missing API key

​

Multi-Turn Conversation Support

OpenRouter agent maintains full conversation history across API calls:

Session Created
  ↓
Load Pending Messages (observations from queue)
  ↓
For each message:
  → Add to conversation history
  → Call OpenRouter API with FULL history
  → Parse XML response
  → Store observations in database
  → Sync to Chroma vector DB
  ↓
Session complete

This enables:

• Coherent multi-turn exchanges
• Context preservation across observations
• Seamless provider switching mid-session

​

Troubleshooting

​

”OpenRouter API key not configured”

Either:

• Set CLAUDE_MEM_OPENROUTER_API_KEY in ~/.claude-mem/settings.json, or
• Set the OPENROUTER_API_KEY environment variable

​

Rate Limiting

Free models may have rate limits during peak usage. If you hit rate limits:

• The agent throws and leaves the message pending — it will be retried later
• Consider switching to a different free model
• Add credits for premium model access

​

Model unavailable

The observer reports OpenRouter model unavailable when OpenRouter says the configured model is retired (“This model has been deprecated”), unknown (“is not a valid model ID”), or has no endpoint left (“No endpoints found for …”). Retrying cannot fix that, so it is not retried; the worker log and the session-start warning tell you to change the model:

• Pick a model from OpenRouter Models and set it as CLAUDE_MEM_OPENROUTER_MODEL in ~/.claude-mem/settings.json (or in the viewer’s Settings). The next observation uses it
• Use the :free suffix for free model variants
• Model IDs are case-sensitive

If you were still on the retired xiaomi/mimo-v2-flash:free default, claude-mem already moved you to cohere/north-mini-code:free the first time it loaded your settings.

​

High Token Usage

If a session accumulates a large conversation history, requests can grow. To keep per-request size down:

• Choose a model suited to your usage volume
• Consider a model with a larger context window if you hit provider-side limits

​

Connection Errors

If you see connection errors:

• Check your internet connection
• Verify OpenRouter service status at status.openrouter.ai
• The agent throws and leaves the message pending for later retry

​

API Details

OpenRouter uses an OpenAI-compatible REST API:

Endpoint: https://openrouter.ai/api/v1/chat/completions

Headers:

Authorization: Bearer {apiKey}
HTTP-Referer: https://github.com/thedotmack/claude-mem
X-Title: claude-mem
Content-Type: application/json

Request Format:

{
  "model": "cohere/north-mini-code:free",
  "messages": [
    {"role": "system", "content": "..."},
    {"role": "user", "content": "..."}
  ],
  "temperature": 0.3,
  "max_tokens": 4096
}

​

Comparing Providers

Feature	Claude (SDK)	Gemini	OpenRouter
Plan usage	Shares your Claude plan	Off-plan (your Gemini key)	Off-plan (your OpenRouter credit or the claude-mem observer)
Models	Claude only	Gemini only	100+ models
Quality	Highest	High	Varies by model
Rate limits	Based on tier	5-4000 RPM	Varies by model
On error	Throws	Throws	Throws
Setup	Automatic	API key required	API key required

Recommendation: The claude-mem observer (chosen at install) is the easiest way to run memory off-plan. Bringing your own key? Start with the free cohere/north-mini-code:free model. If you need higher quality or encounter rate limits, switch to a premium model or the Claude provider.

​

Next Steps

• Configuration - Full settings reference
• Gemini Provider - Alternative free provider
• Getting Started - Basic usage guide
• Troubleshooting - Common issues

​

Multiple keys

CLAUDE_MEM_OPENROUTER_API_KEYS accepts additional keys (newline- or comma-separated) alongside CLAUDE_MEM_OPENROUTER_API_KEY. When a key hits a rate limit, spends its quota, or is rejected as invalid, OpenRouter requests rotate to the next key instead of failing for the rest of the window. See Multi-Key Rotation.