Castellan
Reference · Reeve
Reeve's settings, files and environment
Every setting on Reeve's Settings page with its default, where Reeve keeps its program and data, its ports and scheduled tasks, and the environment variables that change them.
- Article
- 1206
- Applies to
- Reeve 0.17.2
- Last reviewed
- For
- For developers
The Settings page#
Settings, in the title bar of Reeve's page (http://reeve.localhost:18383/#/settings). Each save is checked first, with the same rules Reeve reads its settings by; nothing is written while anything is wrong, and the page says what. A save is refused if the file changed since the page read it.
| Setting | What it does | Default |
|---|---|---|
| Your repositories: Search folders | Where Reeve looks for repositories, directly or one folder down. Filling it in replaces the usual folders. | Empty: the usual folders that exist |
| Your repositories: Other repositories | Repositories anywhere else. | None |
| Your repositories: Left alone | Listed, but no rounds. By path or name. Leave alone on the page adds here. | None |
| The switch for the folders your terminal AI assistant has worked in (Advanced) | Also look after every folder your terminal AI assistant has worked in, wherever it is. | On |
| Delete merged branches on GitHub too (Advanced) | Your own repositories, as owner/name, whose merged branches Reeve may also delete on GitHub: only a branch whose pull request was merged at its tip. | None: this PC only |
| What your AI assistant's model costs per million input tokens | The price Kept off the cloud counts at, from $0.01 to $100. | $3.00 |
| How hard background work may run the graphics card | Work nobody is waiting on (building search indexes, the rounds) rests between requests so a GPU or the CPU is busy at most this much of the time. 100 never rests. Your assistant's own requests always go at full speed, and the NPU is never slowed. | 50% |
| Longest model request (Advanced) | A background chat may take 15 s plus 0.3 s per token, at most this. | 180,000 ms |
| Foundry Local frees its memory after (Advanced) | A Foundry Local model Reeve started is unloaded after this long unused. Only used when no accelerator serves a request. | 60 minutes |
The accelerators aren't here. Which accelerators there are, their order and when idle model servers stop are the Smith's: the page links to the Smith's Settings, or says to hire the Smith. See Where Reeve's work runs.
A running MCP server keeps the settings it started with until your assistant restarts it.
Where Reeve keeps things#
| Installed | Development checkout | |
|---|---|---|
| Program | %USERPROFILE%\.reeve\app | wherever it's cloned |
| Data | %USERPROFILE%\.reeve | %USERPROFILE%\.reeve-dev |
| Page | http://reeve.localhost:18383/ | http://reeve.localhost:28383/ |
| Scheduled tasks | \Reeve\Home page (at sign-in), \Reeve\Rounds (every 30 minutes) | none |
| Settings | %USERPROFILE%\.reeve\config.json | the installed copy's, unless .reeve-dev has its own |
In the data folder:
| What | Where |
|---|---|
| Settings | config.json |
| The digest | digest.md |
| Search indexes | index\<hash of the repository's path> |
| Version-pinned docs | docs\<name>@<version> |
| Reeve's own checkouts | checkouts\<name> |
| Release notes read by the upgrade scout | changelogs\ |
| Jobs' state, approvals and switches | jobs\ |
| Each job run's log | runs\<job>\ |
| Deleted branches, with their tips | branches\deleted.jsonl |
| Kept off the cloud's counts | kept.json |
| Assistants that connected | clients.json |
AI sessions' counts, for ai-use (14 days) | sessions\ |
| Model servers and models it set up without the Smith | servers\, models\ |
The accelerators' list is Castellan's, shared by every agent: %USERPROFILE%\.manor\accelerators\config.json.
In a repository#
A .reeve.json at a repository's top keeps paths out of its search index:
{ "ignore": ["generated/**", "fixtures/**"] }Files .gitignore lists are left out already.
Environment variables#
| Variable | Effect |
|---|---|
REEVE_HOME | The data folder. While it's set, every lock, the shared accelerator files and the servers and models Reeve sets up move under it too, so a scratch home never touches the real ones. |
REEVE_PORT | The page's port. |
REEVE_CONFIG | The config.json to read. |
REEVE_CHAT_URL + REEVE_CHAT_ID (+ REEVE_CHAT_DEVICE) | Chat only from this endpoint (one that speaks the common chat-completions API) and model. |
REEVE_EMBED_URL + REEVE_EMBED_ID (+ REEVE_EMBED_DEVICE) | Embeddings only from this endpoint. REEVE_EMBED_URL=foundry forces Foundry Local. |
REEVE_RERANK_URL + REEVE_RERANK_ID | A reranking server that reorders search's best files, in place of the accelerators' reranker. REEVE_RERANK_FILES (default 10); REEVE_RERANK_BLEND=0 for the reranker's order alone. |
REEVE_TOAST=0 | Reeve raises no Windows notification of its own. |
MANOR_HOME | Castellan's folder, for its theme, its notification settings and your own repositories. Default %USERPROFILE%\.manor. |
Related articles
Is this page right?
If something on it is wrong or out of date, tell us and we'll fix the page.
Still stuck? Write to support@castellan-software.com and mention article 1206. Every version of Reeve, and what changed in it, is in its release notes.