Runs on your own CLI logins · no key to paste · no daily limit

Orchestrate the AI CLIs you already run.

LLMLinq instructs the AI CLIs you already have — whichever you have — with enough precision to get work back you can use. Write better prompts, run a code review, gate a release, and keep every prompt, reply and session where you can find it again.

Install LLMLinq
$ curl -fsSL https://llmlinq.com/install.sh | sh

macOS · Linux · Windows · no Python, no dependencies

An illustration of the LLMLinq timeline: prompts from Claude Code, Codex CLI and Gemini CLI shown together in one list, each with the reply it received.
127.0.0.1:8745today

What LLMLinq does

Five things, all working from one window.

Most of it needs nothing but the install — no account, no upload, no configuration beyond the CLIs you already run.

01 · Connect

One workspace for whichever AI CLI you use

Every AI CLI keeps its own history, in its own format, in its own place. LLMLinq finds whichever ones you have and presents them as one product — so your work stops being scattered across whatever tool happened to do it that day. Nothing to configure, and nothing to install for it.

Found on first run, whichever you have

  • Claude Code · Codex · Antigravity · Gemini · Qwen · Kimi · GLM
  • Ollama, when you would rather run a model yourself
  • Your own API key, when you would rather use that

02 · Work

One page. Every CLI you have.

Write a prompt, choose which tool answers it, and read the reply without leaving the page — Claude Code, Codex, Antigravity, Qwen, Kimi, GLM, or an open model through Ollama. Send the same prompt to a second tool and its answer lands beside the first, so you can see which one understood you. Everything runs under your own login, on your own machine, with no key to paste and no allowance to run out of.

The same question, two tools

  • Pick the tool — the one you pick is the one that runs
  • Generate prompt sharpens what you wrote before you send it
  • Both answers kept, side by side, with what each one took

03 · Review

Two reviewers who never see each other's notes

Point a review at a branch or a pull request and the AI CLIs on your machine read it themselves — they run git, open the files around the change, and judge it against the five dimensions a staff engineer would use. No pasted diffs: a diff cannot answer most of what a review asks, because what decides it usually lives in the code that did not change. With more than one CLI they read at the same time, independently, and anything they both land on is marked — the strongest signal a review of this shape can give you. Tick the findings you agree with and they go to the pull request as inline comments.

What comes back

  • Blocking · security · api.py:42 — with the fix, not just the complaint
  • Both reviewers agreed — the ones to read first
  • Post the ones you agree with, inline, on the PR

04 · Ship

Build the workflow your project actually needs

A workflow is a job you hand to the AI tools you have, worded so the answer comes back usable. The release audit judges a whole project before you tag it — your own tests, linter and audit are run where they exist, and a CLI reads the project against fifteen dimensions with those results as evidence, grading every finding by whether it was verified, needs a tool nobody ran, or was measured from the run. Then build your own: pick which CLIs do the work, paste the prompt you already trust, and save it as a workflow your whole team runs the same way.

Why the grading matters

  • Verified — your test suite ran, and here is what it said
  • Unverified — no scanner ran; it is never rounded up to a pass
  • Ship · ship with conditions · do not ship

05 · Recall

Every prompt and every reply, on one timeline

Every prompt these CLIs recorded, searchable, grouped by day and filtered by project, updating as you work. Each entry carries the answer it received, so months later you can find a decision by what you asked and read the reasoning you actually got — rather than re-explaining the codebase to something that has forgotten it.

Six weeks later

Search "postgres" and the decision comes back — the question you asked, and the reasoning you got.

06 · Ground

A knowledge base every CLI can read

Put your conventions, decisions and runbooks in one place on your machine. Connect it once and your CLIs look things up before they answer, instead of guessing at how your team does things. It travels over MCP, so any CLI that speaks it draws on the same source.

Connect it to Claude Code, once

claude mcp add llmlinq -- llmlinq mcp

It runs on your machine. That is the whole design.

LLMLinq reads your CLI history without changing it, and everything it keeps stays on your machine. Prompts are written by a CLI running under your own login, and reviews happen there too — nothing is uploaded for either. Data reaches our servers only when you look something up in the gallery or press Sync, and both are things you do on purpose.

Read the documentation →

Install in seconds

One command installs a self-contained binary — no Python, no dependencies, no sudo.

curl -fsSL https://llmlinq.com/install.sh | sh

then run llmlinq — your timeline opens at 127.0.0.1:8745

Checksum-verified · re-run to upgrade · all downloads · install docs

Get in touch

A question, a bug, or an idea — it reaches us either way, and we reply by email.