Querying Context¶

Lerim provides several ways to search and retrieve context. All query paths are read-only.

Start with `lerim answer`¶

The main query interface is lerim answer. It queries the shared context DB and returns a synthesized answer grounded in retrieved records.

Recommended agent instruction snippet¶

Add a short instruction to the agent or workflow that should use Lerim context:

## Lerim Context
This workflow uses Lerim for reusable context from prior agent work.
Before repeating investigation, ask Lerim for relevant decisions, constraints,
source evidence, handoffs, and open questions.
Use `lerim answer "your question"` for synthesized answers.
For current runtime state, use `lerim status`.
For detailed usage, invoke the `/lerim` skill.

For coding agents, the same snippet can live in CLAUDE.md, AGENTS.md, or the tool-specific instruction file for the repository.

`lerim answer` -- LLM-powered Q&A¶

The primary query interface. Sends your question through the context answerer with retrieved context records.

Requires running server

lerim answer is a service command that requires lerim up or lerim serve to be running.

Basic query¶

lerim answer "What sources supported our last competitor-pricing assumption?"

The answer flow retrieves relevant records, uses them as context, and returns a natural language answer with evidence of which records were consulted.

Flag	Default	Description
`question`	required	Your question (quote if it contains spaces)
`--scope`	`all`	Read scope: `all` projects or one `project`
`--project`	--	Project name/path when `--scope=project`

JSON output¶

Get structured output for scripting or agent integration:

lerim answer "How is the database configured?" --json

Returns JSON with answer metadata (for example: agent_session_id, error, and cost_usd).

Inspect raw state when needed¶

If you need raw local state instead of a synthesized answer:

Use lerim status for project stream health, queue state, and record counts.
Use lerim query for deterministic record/version/session counts and lists.
Use lerim queue or lerim unscoped for session-ingestion debugging.
Inspect the underlying context store directly only as a last-resort local debugging step. Raw storage access is not the primary agent-facing query path.

Tips for effective queries¶

Be specific¶

# Good -- specific topic
lerim answer "What authentication pattern does the API use?"

# Less effective -- too broad
lerim answer "How does auth work?"

Reference past decisions¶

lerim answer "Why did we switch from REST to gRPC for the internal API?"
lerim answer "What problems did we have with the original caching approach?"

Check before implementing¶

At the start of an agent workflow, ask Lerim for the topic you care about. If you need a synthesized answer across multiple records, use lerim answer:

lerim answer "What evidence supports the latest compliance decision?"