Skip to main content

​
ask

Send a question to the LLM provider with automatic memory context injection. Usage: 
ask <question>
Example: 
> ask What are the benefits of edge AI?
Edge AI provides several key advantages: reduced latency by processing data 
locally, enhanced privacy by keeping sensitive data on-device, lower bandwidth 
costs, and continued operation during network outages. It's ideal for IoT, 
smart home, and industrial applications.
Security: Requires llm authorization Pipeline Flow:
  1. Memory search: Query memory for relevant context (hybrid FTS5 + vector)
  2. Context injection: Add top 5 memory results to prompt
  3. LLM call: Send to configured provider (Anthropic, OpenAI, etc.)
  4. Response: Display LLM output with usage metrics
With Memory Context: 
> remember Edge AI reduces latency by processing data locally
Remembered. (ID: a1b2c3d4, total memories: 5)

> ask Why is edge AI faster?
Internal prompt: 
Related data from memory:
- [02/03 14:30] Edge AI reduces latency by processing data locally

User question: Why is edge AI faster?
LLM response: 
Edge AI is faster because it processes data locally on the device rather than 
sending it to cloud servers. This eliminates network round-trip time, which is 
especially critical for real-time applications like autonomous systems and 
industrial control.

​
Free-Form Text (Implicit LLM)

Any input that doesn’t match a command is automatically sent to the LLM pipeline. Example: 
> Explain vector search
Vector search finds similar items by comparing mathematical representations 
(embeddings) of data. It measures distance between vectors using metrics like 
cosine similarity. This enables semantic search that understands meaning rather 
than just matching keywords.
Equivalent to ask Explain vector search. Security: Requires llm authorization (same as ask)

​
providers

List configured LLM providers and their online status. Usage: 
providers
Example Output: 
LLM Provider:
  anthropic — online
  Chain: anthropic → ollama (fallback)
Security: Requires system:providers authorization Offline Mode: 
> providers
No LLM provider configured (offline mode).
Provider Status:
  • online: API reachable, ready for requests
  • offline: Network error, invalid API key, or service unavailable

​
Provider Selection

OneClaw supports 6 LLM providers:
ProviderModelsUse Case
Anthropicclaude-sonnet-4, claude-opus-4Production, high-quality reasoning
OpenAIgpt-4o, gpt-4o-miniGeneral purpose, function calling
DeepSeekdeepseek-chatCost-effective alternative
Groqllama-3.3-70b, mixtral-8x7bUltra-low latency inference
Googlegemini-2.0-flashMultimodal, large context
Ollamallama3.2, mistral, qwenLocal/offline, privacy-focused

​
Configuration

Primary provider in config/oneclaw.toml: 
[provider]
primary = "anthropic"
model = "claude-sonnet-4-20250514"
max_tokens = 1024
temperature = 0.3
fallback = ["ollama"]  # Try Ollama if Anthropic fails
API keys: 
export ANTHROPIC_API_KEY="sk-ant-..."
export OPENAI_API_KEY="sk-..."
export GOOGLE_API_KEY="AIza..."
# Or set in config/oneclaw.toml [provider.keys]
Ollama (local): 
[provider.ollama]
url = "http://localhost:11434"
model = "llama3.2"

​
Automatic Fallback Chain

When primary provider fails, OneClaw automatically tries fallback providers: Example config: 
[provider]
primary = "anthropic"
fallback = ["openai", "ollama"]
Behavior:
  1. Try Anthropic (primary)
  2. If fails → Try OpenAI (fallback 1)
  3. If fails → Try Ollama (fallback 2)
  4. If all fail → Offline mode response
Offline Mode Response: 
[Offline mode] LLM unavailable (anthropic). Data saved, will process when connected.
2 related entries found in memory.
Verify chain: 
> status
...
Provider: anthropic (online)
Chain: anthropic → openai → ollama (fallback)
...

​
Complexity Analysis

OneClaw automatically analyzes query complexity to optimize LLM usage: Simple queries: Short, direct questions
  • “What is edge AI?”
  • “List the layers”
Complex queries: Multi-step reasoning, memory context
  • “Compare the benefits of edge AI vs cloud AI for industrial IoT”
  • “Based on my stored sensor data, what patterns do you see?”
Impact:
  • Affects model selection in multi-model setups
  • Adjusts context window size
  • Logged in metrics for analysis

​
Context Manager

The orchestrator enriches prompts with:
  1. System prompt: Role definition (“You are OneClaw, a helpful AI assistant…”)
  2. Memory context: Top 5 relevant memory entries
  3. User message: Original query
Example enriched prompt: 
System: You are OneClaw, a helpful AI assistant running on an edge device. 
Answer concisely and clearly. When relevant data is available from memory, 
incorporate it into your response.

Related data from memory:
- [02/03 14:15] The greenhouse temperature threshold is 28°C
- [01/03 09:30] Temperature sensor installed in greenhouse zone B

User: What's the greenhouse temperature limit?

​
Performance Metrics

Track LLM performance via metrics command: 
LLM:
  Calls: 15
  Failed: 1
  Tokens: 12,450
  Latency (total): 3,675ms
  Latency (avg): 245ms
Metrics:
  • Calls: Total LLM API requests
  • Failed: Network errors, rate limits, invalid responses
  • Tokens: Total input + output tokens consumed
  • Latency: Average response time per call
Logging: 
[INFO] complexity=Simple has_memory=true Message analysis
[INFO] provider=anthropic LLM response received

Build docs developers (and LLMs) love

Get started for freeTalk to us