Claude Recall vs. Simon Willison's llm
Simon Willison's llm CLI is the canonical local-LLM logger. SQLite-backed, plugin-friendly, decade of build-in-public. Claude Recall does the same idea for coding-agent CLIs: Claude Code and Codex sessions indexed side by side, with a session-shaped UI on top.
TL;DR
- Use
llmif you talk to many LLMs and want one searchable log. - Use Claude Recall if Claude Code or Codex is your primary tool and you want first-class session-level features. Both are indexed by default.
- Use both if you want a generic LLM log AND an agent-session search UI.
Side-by-side
| Capability | llm CLI | Claude Recall |
|---|---|---|
| SQLite-backed history | Yes | Yes |
| Multi-provider | Yes | Yes, Claude Code and Codex CLI |
| Plugin ecosystem | Yes | Not yet |
| Long-form record of every prompt | Yes | Yes |
| Agent session as a first-class object | No | Yes, Claude Code and Codex |
| Filter any list or search by which agent wrote it | n/a | CLI, web UI and MCP server |
| Pipe a past session into a new conversation | Manual | recall context <id> | claude |
| Web UI / browser timeline | No | Yes |
| Project / tag organization | Light | Yes |
| MCP server | No | Yes |
| Cost analytics per session | No | Yes, at each provider's published list rates |
| Five-plus years of community trust | Yes (11.7K stars) | Small, growing |
Where they win
llm is the right tool if you talk to many models. It captures every prompt across providers into one SQLite database with a clean plugin interface. The community around it (Simon, the Datasette ecosystem) is exceptional. If your use is broad and your favorite model is “whichever one is cheapest today,” llm is the answer.
Where we win
Claude Recall is purpose-built for the coding-agent session as a unit of work. A session has structure (tool calls, file edits, sub-tasks), not just a prompt-response pair. Claude Recall indexes that structure for Claude Code and Codex alike, puts both in one searchable database with a source filter, and pipes any historical session back into a new conversation as condensed context. That is a workflow llm was not built for. The MCP server, per-session cost at each provider's list rates, and the project-aware UI are all consequences of going deep on the session rather than wide on the model catalogue.
When to pick which
- Use
llmfor general-purpose multi-model logging. - Use Claude Recall when a coding-agent CLI is your main interface and you want session-shaped memory. Claude Code and Codex are both indexed, with no configuration for either.
Coexistence
They do not conflict; you can run both. llm will continue to log your llm invocations; Claude Recallwill continue to index your Claude Code and Codex sessions. The two databases live in different places.
FAQ
Is this llm for Claude Code and Codex?
That is the closest one-line analogy and it is how the project is positioned. The implementation is independent. Claude Code and Codex CLI sessions are both indexed, in one database.
Will Claude Recall add multi-provider support?
It already has. Claude Code and Codex CLI sessions sit side by side in one database. Codex indexing is on by default and needs no configuration, the CLI, the web UI and the MCP server each take a source filter, and each session is priced at its own provider's list rates wherever that provider publishes one. Scope discipline is still the moat, it just is not a refusal to index a second agent.
Should I install both?
If you are already an llm user, yes. They are complementary.
Last updated September 2026