Logging, Searching, and Browsing LLM Prompt History

Every prompt you run with LLM is automatically saved — along with the model’s response, token counts, conversation ID, and any attachments or fragments — to a local SQLite database. This lets you search your history, resume conversations, audit costs, and export data for further analysis.

Log Database Location

llm logs path

On macOS this typically outputs:

/Users/simon/Library/Application Support/io.datasette.llm/logs.db

The path varies by operating system and follows the platformdirs convention.

Controlling Logging

Command	Effect
`llm logs off`	Disable logging for all future prompts
`llm logs on`	Re-enable logging
`llm logs status`	Show current status and database statistics
`llm 'prompt' -n` / `--no-log`	Skip logging for a single prompt
`llm 'prompt' --log`	Force logging for a single prompt (when logging is off)

llm logs status

Example output:

Logging is ON for all prompts
Found log database at /Users/simon/Library/Application Support/io.datasette.llm/logs.db
Number of conversations logged: 33
Number of responses logged:     48
Database file size:             19.96MB

Viewing Logs

llm logs

Returns the three most recent logged items in Markdown format. Use -n to change the count:

llm logs -n 10      # ten most recent
llm logs -n 0       # everything ever logged

Output Formats

Flag	Output
(default)	Markdown with prompt and response
`-r` / `--response`	Most recent response only, as plain text
`-x` / `--extract`	First fenced code block from selected entries
`--xl` / `--extract-last`	Last fenced code block
`--json`	Full JSON array
`-t` / `--truncate`	Truncate long prompts and responses
`-u` / `--usage`	Include token usage information

Short Mode

-s / --short prints a compact YAML summary with truncated prompts and no response text — useful for a quick overview:

llm logs -n 2 --short

Example output:

- model: deepseek-reasoner
  datetime: '2025-02-02T06:39:53'
  conversation: 01jk2pk05xq3d0vgk0202zrsg1
  prompt: H01 There are five huts. H02 The Scotsman lives in the purple hut...
- model: o3-mini
  datetime: '2025-02-02T19:03:05'
  conversation: 01jk40qkxetedzpf1zd8k9bgww
  system: Formatting re-enabled. Write a detailed README with extensive usage examples.
  prompt: <documents> <document index="1"> <source>./Cargo.toml</source>...

Add --usage to include token counts:

llm logs -n 1 --short --usage

Filtering Logs

By Conversation

View all logs for the most recent conversation:

llm logs -c

View logs for a specific conversation by ID:

llm logs --cid 01h82n0q9crqtnzmf13gkyxawg

By Search Term

Full-text search across prompts and responses (most relevant first):

llm logs -q 'cheesecake'

Sort by recency instead, and limit results:

llm logs -q 'cheesecake' -l -n 3

By Model

Filter to logs for a specific model or alias:

llm logs -m chatgpt
llm logs -m gpt-4o

Past a Specific ID

Retrieve every log recorded after a known response ID (useful for incremental processing):

llm logs --id-gt 01jm8ec74wxsdatyn5pq1fp0s5
llm logs --id-gte 01jm8ec74wxsdatyn5pq1fp0s5   # inclusive

IDs are issued in ascending time order, so this returns everything since that record.

By Fragment

Filter for logs that used a specific fragment (accepts hash, alias, URL, or file path):

llm logs -f https://llm.datasette.io/robots.txt --expand

Multiple -f options return only responses that used all of the specified fragments.

By Tool

Filter for responses that involved a result from a specific tool:

llm logs -T simple_eval

Show all responses that involved any tool result (including from --functions):

llm logs --tools

By Schema

Filter responses that used a specific schema (any format accepted by --schema):

llm logs --schema 'name, age int, bio'

Combine with --data, --data-array, and --data-key to extract the raw JSON. See the schemas documentation for details.

Browsing Logs with Datasette

Datasette provides a browser-based SQL interface over the logs database:

datasette "$(llm logs path)"

This opens an interactive web UI at http://localhost:8001 where you can run SQL queries, explore tables, and export data.

Backing Up Your Database

llm logs backup /tmp/backup.db

This uses SQLite’s VACUUM INTO to create a clean, compacted copy of the database.

SQL Schema

The logs.db database contains the following tables:

View full SQL schema

CREATE TABLE [conversations] (
  [id] TEXT PRIMARY KEY,
  [name] TEXT,
  [model] TEXT
);
CREATE TABLE [schemas] (
  [id] TEXT PRIMARY KEY,
  [content] TEXT
);
CREATE TABLE "responses" (
  [id] TEXT PRIMARY KEY,
  [model] TEXT,
  [prompt] TEXT,
  [system] TEXT,
  [prompt_json] TEXT,
  [options_json] TEXT,
  [response] TEXT,
  [response_json] TEXT,
  [conversation_id] TEXT REFERENCES [conversations]([id]),
  [duration_ms] INTEGER,
  [datetime_utc] TEXT,
  [input_tokens] INTEGER,
  [output_tokens] INTEGER,
  [token_details] TEXT,
  [schema_id] TEXT REFERENCES [schemas]([id]),
  [resolved_model] TEXT,
  [reasoning] TEXT
);
CREATE VIRTUAL TABLE [responses_fts] USING FTS5 (
  [prompt],
  [response],
  content=[responses]
);
CREATE TABLE [attachments] (
  [id] TEXT PRIMARY KEY,
  [type] TEXT,
  [path] TEXT,
  [url] TEXT,
  [content] BLOB
);
CREATE TABLE [prompt_attachments] (
  [response_id] TEXT REFERENCES [responses]([id]),
  [attachment_id] TEXT REFERENCES [attachments]([id]),
  [order] INTEGER,
  PRIMARY KEY ([response_id], [attachment_id])
);
CREATE TABLE [fragments] (
  [id] INTEGER PRIMARY KEY,
  [hash] TEXT,
  [content] TEXT,
  [datetime_utc] TEXT,
  [source] TEXT
);
CREATE TABLE [fragment_aliases] (
  [alias] TEXT PRIMARY KEY,
  [fragment_id] INTEGER REFERENCES [fragments]([id])
);
CREATE TABLE "prompt_fragments" (
  [response_id] TEXT REFERENCES [responses]([id]),
  [fragment_id] INTEGER REFERENCES [fragments]([id]),
  [order] INTEGER,
  PRIMARY KEY ([response_id], [fragment_id], [order])
);
CREATE TABLE "system_fragments" (
  [response_id] TEXT REFERENCES [responses]([id]),
  [fragment_id] INTEGER REFERENCES [fragments]([id]),
  [order] INTEGER,
  PRIMARY KEY ([response_id], [fragment_id], [order])
);
CREATE TABLE [tools] (
  [id] INTEGER PRIMARY KEY,
  [hash] TEXT,
  [name] TEXT,
  [description] TEXT,
  [input_schema] TEXT,
  [plugin] TEXT
);
CREATE TABLE [tool_responses] (
  [tool_id] INTEGER REFERENCES [tools]([id]),
  [response_id] TEXT REFERENCES [responses]([id]),
  PRIMARY KEY ([tool_id], [response_id])
);
CREATE TABLE [tool_calls] (
  [id] INTEGER PRIMARY KEY,
  [response_id] TEXT REFERENCES [responses]([id]),
  [tool_id] INTEGER REFERENCES [tools]([id]),
  [name] TEXT,
  [arguments] TEXT,
  [tool_call_id] TEXT
);
CREATE TABLE "tool_results" (
  [id] INTEGER PRIMARY KEY,
  [response_id] TEXT REFERENCES [responses]([id]),
  [tool_id] INTEGER REFERENCES [tools]([id]),
  [name] TEXT,
  [output] TEXT,
  [tool_call_id] TEXT,
  [instance_id] INTEGER REFERENCES [tool_instances]([id]),
  [exception] TEXT
);
CREATE TABLE [tool_instances] (
  [id] INTEGER PRIMARY KEY,
  [plugin] TEXT,
  [name] TEXT,
  [arguments] TEXT
);

Key tables

Table	Description
`conversations`	One row per conversation (id, name, model)
`responses`	Every prompt/response pair, linked to a conversation
`responses_fts`	Full-text search index over `prompt` and `response`
`fragments`	Deduplicated fragment content (hash, source, content)
`fragment_aliases`	Named aliases pointing to fragment IDs
`prompt_fragments`	Join table: which fragments were used in which responses
`system_fragments`	Join table: system-prompt fragments per response
`attachments`	Binary or URL-based attachments
`schemas`	JSON schemas referenced by responses
`tools`	Tool definitions used in prompts
`tool_calls`	Individual tool invocations by the model
`tool_results`	Outputs returned by each tool call

responses_fts uses SQLite FTS5 for fast full-text search across prompts and responses.

Get Started

Using LLM

Advanced Features

Embeddings

Plugins

Logging, Searching, and Browsing LLM Prompt History

Log Database Location

Controlling Logging

Viewing Logs

Output Formats

Short Mode

Filtering Logs

By Conversation

By Search Term

By Model

Past a Specific ID

By Fragment

By Tool

By Schema

Browsing Logs with Datasette

Backing Up Your Database

SQL Schema

Key tables

Build docs developers (and LLMs) love

Get Started

Using LLM

Advanced Features

Embeddings

Plugins

Documentation Index

​Log Database Location

​Controlling Logging

​Viewing Logs

​Output Formats

​Short Mode

​Filtering Logs

​By Conversation

​By Search Term

​By Model

​Past a Specific ID

​By Fragment

​By Tool

​By Schema

​Browsing Logs with Datasette

​Backing Up Your Database

​SQL Schema

​Key tables

Build docs developers (and LLMs) love

Log Database Location

Controlling Logging

Viewing Logs

Output Formats

Short Mode

Filtering Logs

By Conversation

By Search Term

By Model

Past a Specific ID

By Fragment

By Tool

By Schema

Browsing Logs with Datasette

Backing Up Your Database

SQL Schema

Key tables