CLI Command Reference

The NeuroLink CLI mirrors the SDK. Every command shares consistent options and outputs so you can prototype in the terminal and port the workflow to code later.

Install or Run Ad-hoc

# Run without installation
npx @juspay/neurolink --help

# Install globally
npm install -g @juspay/neurolink

# Local project dependency
npm install @juspay/neurolink

Command Map

Command	Description	Example
`generate` / `gen`	One-shot content generation with optional multimodal input.	`npx @juspay/neurolink generate "Draft release notes" --image ./before.png`
`stream`	Real-time streaming output with tool support.	`npx @juspay/neurolink stream "Narrate sprint demo" --enableAnalytics`
`batch`	Process multiple prompts from a file.	`npx @juspay/neurolink batch prompts.txt --format json`
`loop`	Interactive session with persistent variables & memory.	`npx @juspay/neurolink loop --auto-redis`
`setup` / `s`	Guided provider onboarding and validation.	`npx @juspay/neurolink setup --provider openai`
`status`	Health check for configured providers.	`npx @juspay/neurolink status --verbose`
`get-best-provider`	Show the best available AI provider.	`npx @juspay/neurolink get-best-provider --format json`
`models list`	Inspect available models and capabilities.	`npx @juspay/neurolink models list --capability vision`
`config <subcommand>`	Initialise, validate, export, or reset configuration.	`npx @juspay/neurolink config validate`
`memory <subcommand>`	View, export, or clear conversation history.	`npx @juspay/neurolink memory history NL_x3yr --format json`
`mcp <subcommand>`	Manage Model Context Protocol servers/tools.	`npx @juspay/neurolink mcp list`
`ollama <subcommand>`	Manage Ollama local AI models.	`npx @juspay/neurolink ollama list-models`
`sagemaker <command>`	Manage Amazon SageMaker endpoints and models.	`npx @juspay/neurolink sagemaker status`
`server <subcommand>`	Manage NeuroLink HTTP server
`serve`	Start server in foreground mode
`validate`	Alias for `config validate`.	`npx @juspay/neurolink validate`
`completion`	Generate shell completion script.	`npx @juspay/neurolink completion > ~/.neurolink-completion.sh`

Primary Commands

`generate <input>`

npx @juspay/neurolink generate "Summarise design doc" \
  --provider google-ai --model gemini-2.5-pro \
  --image ./screenshots/ui.png --enableAnalytics --enableEvaluation

Key flags:

--provider, -p – provider slug (default auto).
--model, -m – model name for the chosen provider.
--image, -i – attach one or more image files/URLs for multimodal prompts.
--pdf – attach one or more PDF files for document analysis.
--csv, -c – attach one or more CSV files for data analysis.
--file – attach any supported file type (auto-detected: Excel, Word, RTF, JSON, YAML, XML, HTML, SVG, Markdown, code files, and more).
--temperature, -t – creativity (default 0.7).
--maxTokens, --max – response limit (default 1000).
--system, -s – system prompt.
--format, -f, --output-format – text (default), json, or table.
--output, -o – write response to file.
--imageOutput, --image-output – custom path for generated image (default: generated-images/image-<timestamp>.png).
--enableAnalytics / --enableEvaluation – capture metrics & quality scores.
--evaluationDomain – domain hint for the judge model.
--domainAware – use domain-aware evaluation (default false).
--context – JSON string appended to analytics/evaluation context.
--domain, -d – domain type for specialized processing: healthcare, finance, analytics, ecommerce, education, legal, technology, generic, auto.
--disableTools – bypass MCP tools for this call.
--timeout – seconds before aborting the request (default 120).
--region, -r – Vertex AI region (e.g., us-central1, europe-west1, asia-northeast1).
--debug, -v, --verbose – verbose logging and full JSON payloads.
--quiet, -q – suppress non-essential output (default true).

CSV Options:

--csvMaxRows – maximum number of CSV rows to process (default 1000).
--csvFormat – CSV output format: raw (default), markdown, json.

Video Input (Analysis):

--video – attach video file for analysis (MP4, WebM, MOV, AVI, MKV).
--video-frames – number of frames to extract (default 8).
--video-quality – frame quality 0–100 (default 85).
--video-format – frame format: jpeg (default) or png.
--transcribe-audio – extract and transcribe audio from video (default false).

Text-to-Speech (TTS):

--tts – enable text-to-speech output (default false).
--ttsVoice – TTS voice to use (e.g., en-US-Neural2-C).
--ttsFormat – audio output format: mp3 (default), wav, ogg, opus.
--ttsSpeed – speaking rate 0.25–4.0 (default 1.0).
--ttsQuality – audio quality level: standard (default) or hd.
--ttsOutput – save TTS audio to file (supports absolute and relative paths).
--ttsPlay – auto-play generated audio (default false).

Extended Thinking:

--thinking, --think – enable extended thinking/reasoning capability (default false).
--thinkingBudget – token budget for extended thinking (5000–100000, default 10000). Supported by Anthropic Claude and Gemini 2.5+ models.
--thinkingLevel – thinking level for Gemini 3 models: minimal, low, medium, high.

File Input Examples:

# Attach multiple file types
npx @juspay/neurolink generate "Analyze this data" \
  --file ./report.xlsx \
  --file ./config.yaml \
  --file ./diagram.svg

# Mix file types with images and PDFs
npx @juspay/neurolink generate "Compare architecture" \
  --file ./main.ts \
  --pdf ./spec.pdf \
  --image ./screenshot.png

See File Processors Guide for all 17+ supported file types.

Video Generation (Veo 3.1):

--outputMode – output mode: text (default) or video.
--image – path to input image file (required for video generation, e.g., ./input.jpg).
--videoOutput, -vo – path to save generated video file.
--videoResolution – 720p or 1080p (default 720p).
--videoLength – duration: 4, 6, or 8 seconds (default 4).
--videoAspectRatio – 9:16 (portrait) or 16:9 (landscape, default 16:9).
--videoAudio – include synchronized audio (default true).

Note: Video generation requires Vertex AI provider (vertex) and Veo 3.1 model (veo-3.1). The provider auto-switches to Vertex when --outputMode video is specified. Supported image formats: PNG, JPEG, WebP (max 20MB).

gen is a short alias with the same options.

`stream <input>`

npx @juspay/neurolink stream "Walk through the timeline" \
  --provider openai --model gpt-4o --enableEvaluation

stream shares the same flags as generate and adds chunked output for live UIs. Evaluation results are emitted after the stream completes when --enableEvaluation is set.

`batch <file>`

Process multiple prompts from a file in sequence.

# Process prompts from a file
npx @juspay/neurolink batch prompts.txt

# Export results as JSON
npx @juspay/neurolink batch questions.txt --format json

# Use Vertex AI with 2s delay between requests
npx @juspay/neurolink batch tasks.txt -p vertex --delay 2000

# Save results to file
npx @juspay/neurolink batch batch.txt --output results.json

batch shares the same flags as generate. The input file should contain one prompt per line. Results are returned as an array of { prompt, response } objects. A default 1-second delay is applied between requests; override with --delay <ms>.

Model Evaluation

Evaluate AI model outputs for quality, accuracy, and safety using NeuroLink's built-in evaluation engine.

Via generate/stream commands:

# Enable evaluation on any command
npx @juspay/neurolink generate "Write a product description" \
  --enableEvaluation \
  --evaluationDomain "e-commerce"

Evaluation Output:

{
  "response": "...",
  "evaluation": {
    "score": 0.85,
    "metrics": {
      "accuracy": 0.9,
      "safety": 1.0,
      "relevance": 0.8
    },
    "judge_model": "gpt-4o",
    "feedback": "High quality response with clear structure"
  }
}

Key Evaluation Flags:

--enableEvaluation – Activate quality scoring
--evaluationDomain <domain> – Context hint for the judge (e.g., "medical", "legal", "technical")
--context <json> – Additional context for evaluation

Judge Models:

NeuroLink uses GPT-4o by default as the judge model, but you can configure different models for evaluation in your SDK configuration.

Use Cases:

Quality assurance for production outputs
A/B testing different prompts
Safety validation before deployment
Compliance checking for regulated industries

Learn more: Auto Evaluation Guide

`loop`

Interactive session mode with persistent state, conversation memory, and session variables. Perfect for iterative workflows and experimentation.

# Start loop with Redis-backed conversation memory
npx @juspay/neurolink loop --enable-conversation-memory --auto-redis

# Start loop without Redis auto-detection
npx @juspay/neurolink loop --enable-conversation-memory --no-auto-redis

# Force start a new conversation (skip selection menu)
npx @juspay/neurolink loop --new

# Resume a specific conversation by session ID
npx @juspay/neurolink loop --resume abc123def456

# List available conversations and exit
npx @juspay/neurolink loop --list-conversations

# Use in-memory storage only
npx @juspay/neurolink loop --no-auto-redis

Loop-specific flags:

Flag	Alias	Type	Default	Description
`--enable-conversation-memory`		boolean	true	Enable conversation memory for the loop session
`--max-sessions`		number	50	Maximum number of conversation sessions to keep
`--max-turns-per-session`		number	20	Maximum turns per conversation session
`--auto-redis`		boolean	true	Automatically use Redis if available
`--resume`	`-r`	string		Directly resume a specific conversation by session ID
`--new`	`-n`	boolean		Force start a new conversation (skip selection menu)
`--list-conversations`	`-l`	boolean		List available conversations and exit
`--compact-threshold`		number	0.8	Context compaction trigger threshold (0.0–1.0)
`--disable-compaction`		boolean	false	Disable automatic context compaction

Key capabilities:

Run any CLI command without restarting session
Persistent session variables: set provider openai, set temperature 0.9
Conversation memory: AI remembers previous turns within session
Redis auto-detection: Automatically connects if REDIS_URL is set
Export session history as JSON for analytics
Automatic context compaction when usage exceeds threshold

Session management commands (inside loop):

Command	Description
`help`	Show all available loop mode commands and standard CLI help.
`set <key> <value>`	Set a session variable. Use `set help` for available keys.
`get <key>`	Show current value of a session variable.
`unset <key>`	Remove a session variable.
`show`	Display all currently set session variables.
`clear`	Reset all session variables.
`exit`	Exit loop session. Aliases: `quit`, `:q`.

Settable session variables (via set):

Variable	Type	Description	Allowed Values
`provider`	string	The AI provider to use.	`openai`, `anthropic`, `google-ai`, `vertex`, `bedrock`, `azure`, etc.
`model`	string	The specific model to use from the provider.	Any valid model name
`temperature`	number	Controls randomness of the output (e.g., 0.2, 0.8).
`maxTokens`	number	The maximum number of tokens to generate.
`output`	string	AI response format value.	`text`, `json`, `structured`, `none`
`systemPrompt`	string	The system prompt to guide the AI's behavior.
`timeout`	number	Timeout for the generation request in milliseconds.
`disableTools`	boolean	Disable all tool usage for the AI.
`maxSteps`	number	Maximum number of tool execution steps.
`enableAnalytics`	boolean	Enable or disable analytics for responses.
`enableEvaluation`	boolean	Enable or disable AI-powered evaluation of responses.
`evaluationDomain`	string	Domain expertise for evaluation.
`toolUsageContext`	string	Context about tools/MCPs used in the interaction.
`enableSummarization`	boolean	Enable automatic conversation summarization.
`thinking`	boolean	Enable extended thinking/reasoning capability.
`thinkingBudget`	number	Token budget for thinking (Anthropic models: 5000–100000).
`thinkingLevel`	string	Thinking level for Gemini 3 models.	`minimal`, `low`, `medium`, `high`

Context Budget Warnings:

During a loop session, NeuroLink monitors context window usage after each generation command:

60% used (gray): A subtle status line is shown: Context: 62% used.
80% used (yellow): A prominent warning with token counts is shown:
```
Context usage: 83% of window (12,450 / 15,000 tokens)
Auto-compaction will trigger to preserve conversation quality.
```
When --disable-compaction is not set, the system automatically compacts the context to free up space while preserving conversation quality.

See the complete guide: CLI Loop Sessions

`setup`

Interactive provider configuration wizard that guides you through API key setup, credential validation, and recommended model selection.

# Launch interactive setup wizard
npx @juspay/neurolink setup

# Show all available providers
npx @juspay/neurolink setup --list

# Configure a specific provider
npx @juspay/neurolink setup --provider openai
npx @juspay/neurolink setup --provider bedrock
npx @juspay/neurolink setup --provider google-ai

What the wizard does:

Prompts for API keys – Securely collects credentials
Validates authentication – Tests connection to provider
Writes .env file – Safely stores credentials (creates if missing)
Recommends models – Suggests best models for your use case
Shows example commands – Quick-start examples to try immediately

Supported providers: OpenAI, Anthropic, Google AI, Vertex AI, Bedrock, Azure, Hugging Face, Ollama, Mistral, and more.

`status`

npx @juspay/neurolink status --verbose

Displays provider availability, authentication status, recent error summaries, and response latency.

`models`

# List all models for a provider
npx @juspay/neurolink models list --provider google-ai

# Filter by capability
npx @juspay/neurolink models list --capability vision --format table

`config`

Manage persistent configuration stored in the NeuroLink config directory.

npx @juspay/neurolink config init
npx @juspay/neurolink config validate
npx @juspay/neurolink config export --format json > neurolink-config.json

`memory`

Manage conversation history stored in Redis. View, export, or clear session data for analytics and debugging.

# List all active sessions
npx @juspay/neurolink memory list

# View session statistics
npx @juspay/neurolink memory stats

# View conversation history (text format)
npx @juspay/neurolink memory history <SESSION_ID>

# Export session as JSON (Q4 2025 - for analytics)
npx @juspay/neurolink memory export --session-id <SESSION_ID> --format json > session.json

# Export all sessions
npx @juspay/neurolink memory export-all --output ./exports/

# Delete a single session
npx @juspay/neurolink memory clear <SESSION_ID>

# Delete all sessions
npx @juspay/neurolink memory clear-all

Export formats:

json – Structured data with metadata, timestamps, token counts
csv – Tabular format for spreadsheet analysis

Note: Requires Redis-backed conversation memory. Set REDIS_URL environment variable.

See the complete guide: Redis Conversation Export

`mcp`

Manage Model Context Protocol servers and tools. Supports stdio, SSE, WebSocket, and HTTP transports.

# List registered servers/tools
npx @juspay/neurolink mcp list

# Auto-discover MCP servers from config files
npx @juspay/neurolink mcp discover

# Install popular MCP servers
npx @juspay/neurolink mcp install filesystem
npx @juspay/neurolink mcp install github

# Add custom servers with different transports
npx @juspay/neurolink mcp add myserver "python server.py" --transport stdio
npx @juspay/neurolink mcp add webserver "http://localhost:8080" --transport sse --url "http://localhost:8080/sse"

# Add HTTP remote server with authentication
npx @juspay/neurolink mcp add remote-api "https://api.example.com/mcp" \
  --transport http \
  --url "https://api.example.com/mcp" \
  --headers '{"Authorization": "Bearer YOUR_TOKEN"}'

# Test server connectivity
npx @juspay/neurolink mcp test myserver

# Remove a server
npx @juspay/neurolink mcp remove myserver

MCP Command Options:

Option	Description
`--transport`	Transport type: `stdio`, `sse`, `websocket`, `http`
`--url`	URL for SSE/WebSocket/HTTP transport
`--headers`	JSON string with HTTP headers for authentication
`--args`	Command arguments (comma-separated)
`--env`	Environment variables (JSON string)
`--cwd`	Working directory for the server

HTTP Transport Features:

Custom headers for authentication (Bearer tokens, API keys)
Configurable timeouts and connection options
Automatic retry with exponential backoff
Rate limiting to prevent API throttling
OAuth 2.1 support with PKCE

See MCP HTTP Transport Guide for complete configuration options.

`batch`

See batch <file> above.

`get-best-provider`

Show the best available AI provider based on current configuration and availability.

# Get best available provider
npx @juspay/neurolink get-best-provider

# Get provider as JSON
npx @juspay/neurolink get-best-provider --format json

# Just the provider name
npx @juspay/neurolink get-best-provider --quiet

`ollama <command>`

Manage Ollama local AI models. Requires Ollama to be installed on the local machine.

# List installed models
npx @juspay/neurolink ollama list-models

# Download a model
npx @juspay/neurolink ollama pull llama3

# Remove a model
npx @juspay/neurolink ollama remove llama3

# Check Ollama service status
npx @juspay/neurolink ollama status

# Start/stop Ollama service
npx @juspay/neurolink ollama start
npx @juspay/neurolink ollama stop

# Interactive Ollama setup
npx @juspay/neurolink ollama setup

Subcommands:

Subcommand	Description
`list-models`	List installed Ollama models
`pull <model>`	Download an Ollama model
`remove <model>`	Remove an Ollama model
`status`	Check Ollama service status
`start`	Start Ollama service
`stop`	Stop Ollama service
`setup`	Interactive Ollama setup

`sagemaker <command>`

Manage Amazon SageMaker AI models and endpoints.

# Check SageMaker configuration and connectivity
npx @juspay/neurolink sagemaker status

# Test connectivity to an endpoint
npx @juspay/neurolink sagemaker test my-endpoint

# List available endpoints
npx @juspay/neurolink sagemaker list-endpoints

# Show current SageMaker configuration
npx @juspay/neurolink sagemaker config

# Interactive setup
npx @juspay/neurolink sagemaker setup

# Validate configuration and credentials
npx @juspay/neurolink sagemaker validate

# Run performance benchmark
npx @juspay/neurolink sagemaker benchmark my-endpoint

Subcommands:

Subcommand	Description
`status`	Check SageMaker configuration and connectivity
`test <endpoint>`	Test connectivity to a SageMaker endpoint
`list-endpoints`	List available SageMaker endpoints
`config`	Show current SageMaker configuration
`setup`	Interactive SageMaker configuration setup
`validate`	Validate SageMaker configuration and credentials
`benchmark <endpoint>`	Run performance benchmark against endpoint

`completion`

Generate a shell completion script for bash.

# Generate shell completion
npx @juspay/neurolink completion

# Save completion script
npx @juspay/neurolink completion > ~/.neurolink-completion.sh

# Enable completions (bash)
source ~/.neurolink-completion.sh

Add the completion script to your shell profile for persistent completions.

`serve`

Start the NeuroLink HTTP server in foreground mode.

Usage

neurolink serve [options]

Options

Option	Alias	Type	Default	Description
`--port`	`-p`	number	3000	Port to listen on
`--host`	`-H`	string	0.0.0.0	Host to bind to
`--framework`	`-f`	string	hono	Web framework: hono, express, fastify, koa
`--basePath`		string	/api	Base path for all routes
`--cors`		boolean	true	Enable CORS
`--rateLimit`		number	100	Rate limit (requests per 15-minute window, 0 to disable)
`--swagger`		boolean	false	Enable Swagger UI and OpenAPI endpoints
`--watch`	`-w`	boolean	false	Enable watch mode
`--config`	`-c`	string		Path to config file

Swagger/OpenAPI Endpoints

When --swagger is enabled, these endpoints become available:

Endpoint	Description
`GET /api/openapi.json`	OpenAPI 3.1 specification in JSON format
`GET /api/openapi.yaml`	OpenAPI 3.1 specification in YAML format
`GET /api/docs`	Interactive Swagger UI documentation

Note: Disable with --no-swagger in production to avoid exposing API structure.

Examples

# Start with defaults
neurolink serve

# Start on specific port with Express
neurolink serve --port 8080 --framework express

# Start with custom config file
neurolink serve --config ./server.config.json

`server <subcommand>`

Manage NeuroLink HTTP server for exposing AI agents as REST APIs.

Subcommands

Subcommand	Description
`start`	Start the HTTP server in background
`stop`	Stop the running server
`status`	Show server status
`routes`	List all registered routes
`config`	Show or modify server configuration
`openapi`	Generate OpenAPI specification

`server start`

Start the HTTP server in background mode.

neurolink server start [options]

Option	Alias	Type	Default	Description
`--port`	`-p`	number	3000	Port to listen on
`--host`	`-H`	string	0.0.0.0	Host to bind to
`--framework`	`-f`	string	hono	Framework: hono, express, fastify, koa
`--basePath`		string	/api	Base path for all routes
`--cors`		boolean	true	Enable CORS
`--rateLimit`		number	100	Rate limit (requests per 15-minute window, 0 to disable)

Examples:

# Start with defaults
neurolink server start

# Start on port 8080 with Express
neurolink server start -p 8080 --framework express

`server stop`

Stop a running background server.

neurolink server stop [options]

Option	Type	Default	Description
`--force`	boolean	false	Force stop even if server is not responding

Examples:

# Stop gracefully
neurolink server stop

# Force stop
neurolink server stop --force

`server status`

Show server status information.

neurolink server status [options]

Option	Type	Default	Description
`--format`	string	text	Output format: text, json

Examples:

# Text output
neurolink server status

# JSON output for scripting
neurolink server status --format json

`server routes`

List all registered server routes.

neurolink server routes [options]

Option	Type	Default	Description
`--format`	string	table	Output format: text, json, table
`--group`	string	all	Filter by route group: agent, tool, mcp, memory, health, all
`--method`	string	all	Filter by HTTP method: GET, POST, PUT, DELETE, PATCH, all

Examples:

# List all routes in table format
neurolink server routes

# List only agent routes
neurolink server routes --group agent

# List all POST endpoints as JSON
neurolink server routes --method POST --format json

`server config`

Show or modify server configuration.

neurolink server config [options]

Option	Type	Default	Description
`--get`	string		Get a specific config value
`--set`	string		Set a config value (format: key=value)
`--reset`	boolean	false	Reset configuration to defaults
`--format`	string	text	Output format: text, json

Examples:

# Show all configuration
neurolink server config

# Get specific value
neurolink server config --get defaultPort

# Set a value
neurolink server config --set defaultPort=8080

# Reset to defaults
neurolink server config --reset

`server openapi`

Generate OpenAPI specification.

neurolink server openapi [options]

Option	Alias	Type	Default	Description
`--output`	`-o`	string	stdout	Output file path
`--format`		string	json	Output format: json, yaml
`--basePath`		string	/api	Base path for all routes
`--title`		string		API title
`--version`		string		API version

Examples:

# Generate to stdout
neurolink server openapi

# Save to file
neurolink server openapi -o openapi.json

# Generate YAML format
neurolink server openapi --format yaml -o openapi.yaml

Global Flags (available on every command)

Flag	Alias	Default	Description
`--provider`	`-p`	`auto`	AI provider to use (auto-selects best available).
`--model`	`-m`		Specific model to use.
`--temperature`	`-t`	`0.7`	Creativity level (0.0 = focused, 1.0 = creative).
`--maxTokens`	`--max`	`1000`	Maximum tokens to generate.
`--system`	`-s`		System prompt to guide AI behavior.
`--format`	`-f`, `--output-format`	`text`	Output format: `text`, `json`, `table`.
`--output`	`-o`		Save output to file.
`--configFile <path>`			Use a specific configuration file.
`--dryRun`		`false`	Generate without calling providers (returns mocked analytics/evaluation).
`--noColor`		`false`	Disable ANSI colours.
`--delay <ms>`			Delay between batched operations.
`--domain <slug>`	`-d`		Domain type for specialized processing and optimization.
`--toolUsageContext <text>`			Describe expected tool usage for better evaluation feedback.
`--debug`	`-v`, `--verbose`	`false`	Enable debug mode with verbose output.
`--quiet`	`-q`	`true`	Suppress non-essential output.
`--timeout`		`120`	Maximum execution time in seconds.
`--disableTools`		`false`	Disable MCP tool integration.
`--enableAnalytics`		`false`	Enable usage analytics collection.
`--enableEvaluation`		`false`	Enable AI response quality evaluation.
`--region`	`-r`		Vertex AI region (e.g., `us-central1`).

JSON-Friendly Automation

--format json returns structured output including analytics, evaluation, tool calls, and response metadata.
Combine with --enableAnalytics --enableEvaluation to capture usage costs and quality scores in automation pipelines.
Use --output <file> to persist raw responses alongside JSON logs.

rag <subcommand>

Document processing and RAG pipeline commands.

Subcommand	Description
`chunk`	Chunk a document using a specified strategy
`index`	Index documents into a vector store
`query`	Query indexed documents

rag chunk

Chunk a document file into smaller pieces for RAG processing.

neurolink rag chunk <file> [options]

Option	Alias	Type	Default	Description
`--strategy`	`-s`	string	`recursive`	Chunking strategy
`--chunk-size`		number	`1000`	Maximum chunk size
`--chunk-overlap`		number	`200`	Overlap between chunks
`--output`	`-o`	string	stdout	Output file path
`--format`	`-f`	string	`text`	Output format (text, json)

Chunking Strategies: character, recursive, sentence, token, markdown, html, json, latex, semantic, semantic-markdown

Examples:

# Default chunking
neurolink rag chunk ./docs/guide.md

# Markdown-aware chunking with JSON output
neurolink rag chunk ./docs/guide.md --strategy markdown --format json

# Custom size and overlap
neurolink rag chunk ./docs/guide.md --chunk-size 512 --chunk-overlap 50 --output chunks.json

RAG Flags on generate/stream

RAG can also be used directly with generate and stream commands via --rag-files:

neurolink generate "What is this about?" --rag-files ./docs/guide.md
neurolink stream "Summarize" --rag-files ./docs/a.md ./docs/b.md --rag-top-k 10

Flag	Type	Default	Description
`--rag-files`	string[]	-	File paths to load for RAG context
`--rag-strategy`	string	auto-detected	Chunking strategy for RAG documents
`--rag-chunk-size`	number	1000	Maximum chunk size in characters
`--rag-chunk-overlap`	number	200	Overlap between adjacent chunks
`--rag-top-k`	number	5	Number of top results to retrieve

Troubleshooting

Issue	Tip
`Unknown argument`	Check spelling; run `command --help` for the latest options.
CLI exits immediately	Upgrade to the newest release or clear old `neurolink` binaries on PATH.
Provider shows as `not-configured`	Run `neurolink setup --provider <name>` or populate `.env`.
Analytics/evaluation missing	Ensure both `--enableAnalytics`/`--enableEvaluation` and provider credentials for the judge model exist.

For advanced workflows (batching, tooling, configuration management) see the relevant guides in the documentation sidebar.

Q4 2025:

CLI Loop Sessions – Persistent interactive mode with session management
Redis Conversation Export – Export session history via memory export
Guardrails Middleware – Content filtering (use --middleware-preset security)

Q3 2025:

Multimodal Chat – Use --image flag with generate or stream
Auto Evaluation – Enable with --enableEvaluation
Provider Orchestration – Automatic fallback and routing

Documentation:

SDK API Reference – TypeScript API equivalents
Configuration Guide – Environment variables and config files
Troubleshooting – Detailed error solutions

Install or Run Ad-hoc​

Command Map​

Primary Commands​

generate <input>​

stream <input>​

batch <file>​

Model Evaluation​

loop​

setup​

status​

models​

config​

memory​

mcp​

batch​

get-best-provider​

ollama <command>​

sagemaker <command>​

completion​

serve​

Usage​

Options​

Swagger/OpenAPI Endpoints​

Examples​

server <subcommand>​

Subcommands​

server start​

server stop​

server status​

server routes​

server config​

server openapi​

Global Flags (available on every command)​

JSON-Friendly Automation​

rag <subcommand>​

rag chunk​

RAG Flags on generate/stream​

Troubleshooting​

Related Features​

Install or Run Ad-hoc

Command Map

Primary Commands

`generate <input>`

`stream <input>`

`batch <file>`

Model Evaluation

`loop`

`setup`

`status`

`models`

`config`

`memory`

`mcp`

`batch`

`get-best-provider`

`ollama <command>`

`sagemaker <command>`

`completion`

`serve`

Usage

Options

Swagger/OpenAPI Endpoints

Examples

`server <subcommand>`

Subcommands

`server start`

`server stop`

`server status`

`server routes`

`server config`

`server openapi`

Global Flags (available on every command)

JSON-Friendly Automation

rag <subcommand>

rag chunk

RAG Flags on generate/stream

Troubleshooting

Related Features