Skip to main content

​
Overview

One-shot query: ask Lerim a question and get an answer informed by memories extracted from your agent sessions.
This command requires a running Lerim server. Start it with lerim up (Docker) or lerim serve (direct).

​
Syntax

lerim ask <question> [--limit N] [--project NAME] [--json]

​
Parameters

​
question
string
required
Your question (use quotes if it contains spaces).
​
--limit
integer
default:"12"
Maximum number of memory items to include as context.
​
--project
string
Scope to a specific project. Note: Not yet implemented.
​
--json
boolean
default:"false"
Output structured JSON instead of human-readable text.

​
Examples

​
Basic question

Ask about authentication patterns: 
lerim ask 'What auth pattern do we use?'
Output: 
Based on your project memories, you use bearer token authentication
for API requests. This pattern was chosen for its simplicity and
compatibility with standard HTTP clients.

​
Limited context

Limit the number of memories used as context: 
lerim ask "How is the database configured?" --limit 5

​
JSON output

Get structured output for parsing: 
lerim ask "What testing framework do we use?" --json
Output: 
{
  "question": "What testing framework do we use?",
  "answer": "Your project uses pytest as the primary testing framework...",
  "memories_used": 8,
  "confidence": 0.87
}

​
How it works

  1. Your question is sent to the running Lerim server via HTTP POST to /api/ask
  2. Lerim performs memory retrieval to find relevant decisions and learnings
  3. The top N memories (default 12) are used as context
  4. An LLM generates an answer informed by these memories
  5. The answer is returned with citation evidence
For best results, ask specific questions about decisions, patterns, or procedures in your project.

​
Exit codes

  • 0: Success - answer generated
  • 1: Error - server not running or authentication failed
  • 2: Usage error - invalid arguments

lerim memory search

Search memories by keyword

lerim status

Check server status

​
Notes

  • Ask uses memory retrieval evidence to ground its answers
  • If provider auth fails (missing API key), the CLI returns exit code 1
  • The --project flag is reserved for future project-scoped queries

Build docs developers (and LLMs) love

Get started for freeTalk to us