Get Rema answering from what your team wrote

npx askrema@latest

cd ~/rema-demo/ml-service && claude

hey rema, our pyproject sets index-strategy = unsafe-any-match so torch comes from the pytorch index. is that still right, and is it safe?

cd <your project>
npx askrema ingest ./docs

npx askrema@latest

~/.rema/bin/askrema doctor      # why the "Rema found" line did not appear
~/.rema/bin/askrema status      # what is installed and where
~/.rema/bin/askrema uninstall   # remove what the install added, and nothing else

# Set up Rema and get a first cited answer

Rema puts what a team already wrote (issues, review threads, docs) in front of the coding agent,
with a link for every passage. You are setting it up here. Do the work yourself: run the commands
and check what they print. Do not just explain the steps.

## 1. Install
- Run exactly: npx -y askrema@latest --yes
- The installer prints its plan (every file it will create or change) and normally asks once. The
  trailing --yes answers that question, because your shell has no terminal to ask in. Show me the
  plan it printed.
- It needs Node 18 or newer on macOS or Linux. It does not need sudo, an API key or Python: if
  there is no Python 3.11+, it downloads a private copy into ~/.rema.
- The package name is askrema. Do not substitute "npx rema", "npm install rema" or
  "pip install rema": those are other people's packages.
- Expected: lines starting "ok" (python, cli, one hook line per agent, demo, check), then
  "Installed in ...s". The "check" line starts "Rema found 11 passages".

## 2. Tell me what changed on my machine
- ~/.rema holds the CLI. ~/rema-demo holds a demo project built from public astral-sh/uv history.
- One UserPromptSubmit hook was added to ~/.claude/settings.json (and to ~/.codex/hooks.json if I
  use Codex, which asks me to trust the hook: /hooks). A copy of any settings file it changed is
  saved next to it as <name>.before-rema. The hook reads .rema/evidence.db in the project I am in
  and does nothing in a project that has none. Rema itself sends nothing anywhere; the passages it
  adds become part of my prompt, so they reach my agent's model provider like anything I type.
- "npx askrema uninstall" removes all of it.

## 3. Hand over the first question
The hook loads when a session starts, so it is not active in this session. Tell me to open a new
terminal and run:

  cd ~/rema-demo/ml-service && claude

and then say:

  hey rema, our pyproject sets index-strategy = unsafe-any-match so torch comes from the pytorch index. is that still right, and is it safe?

Under the question there should be a line starting "Rema found", then an answer that cites the uv
threads it came from. That is the whole interface: in any project with a Rema store, start a
prompt with "hey rema," and ask as usual. There is no command to learn.

## 4. My own project (ask me first)
- Ask me which folder holds what my team wrote (docs, exported threads, design notes).
- From the root of my project, run: npx askrema ingest <that folder>
  It prints one result per file. A file the injection gate refuses is listed with the reason: show
  me those, and do not pass --allow-unsafe yourself.
- The store is .rema/evidence.db in the project. In a new session there, say "hey rema, ..." and
  ask as usual.

## 5. Report
- What the installer printed, which files it changed, and anything that failed, with the exact
  command and error.

Rules: no step needs an API key, so do not set, read or copy one. Do not send any file anywhere.
Do not edit my shell profile. If a step fails, stop and report it. Do not swap in a different
command to make it look like it worked.

mkdir rema-try && cd rema-try
curl -sL -o CHANGELOG.md https://raw.githubusercontent.com/astral-sh/uv/bea138450f0e620a4ce5765b0e38cff7b9f0799f/CHANGELOG.md
rema ingest CHANGELOG.md

"path": "CHANGELOG.md",
"accepted": true,
"chunks_written": 133,

rema search --limit 1 "which preview feature runs build backend hooks with lazy imports"

"health": "ok",
"no_match": false,
"hits": [
  {
    "rank": 1,
    "locator": "section:Preview features@19-23",
    "text": "- Run build-backend hooks with lazy imports on CPython 3.15 and later using the `build-lazy-imports` preview feature ..."
  }
]

rema search "what is the wifi password"

"hits": [],
"no_match": true,
"detail": "no chunk matches the content words of this question (wifi, password), and the control query 'https' did return a hit, so the index is answering. This is a real absence, not a broken index."

mkdir -p sessions
cat > sessions/retry-backoff.jsonl <<'EOF'
{"role": "user", "content": "the nightly sync is slow. drop the retry backoff in billing/retry.py from 30s to 5s"}
{"role": "assistant", "content": "Changed BACKOFF_SECONDS from 30 to 5.", "tool_name": "Edit", "file_path": "billing/retry.py"}
{"role": "user", "content": "no, put it back to 30. we tried 5s in march and the payment provider rate limited us for an hour. the slow part is the page size, not the backoff."}
{"role": "assistant", "content": "Reverted BACKOFF_SECONDS to 30.", "tool_name": "Edit", "file_path": "billing/retry.py"}
EOF
rema capture sessions

rema why billing/retry.py --question "can we lower the backoff"

"answered": true,
"reason": "answered from 1 recorded session(s)",
"citations": [
  {
    "quote": "no, put it back to 30. we tried 5s in march and the payment provider rate limited us for an hour. the slow part is the page size, not the backoff.",
    "locator": "session:retry-backoff!turn:3",
    "rank": 1
  },
  ...

rema capture --dry-run      # what it would read, nothing stored
rema capture                # read it
rema why billing/retry.py --question "can we lower the backoff"