Skip to main content
Glama
derwells

sieve

by derwells

Server Configuration

Describes the environment variables required to run the server.

NameRequiredDescriptionDefault
BRAVE_API_KEYNoOptional. Your Brave Search API key, required if using the 'brave' search backend.
SIEVE_ENV_FILENoOptional. Path to an env file that the launcher sources. Defaults to ~/.config/sieve/env.
SIEVE_SEARCH_CMDNoOptional. Overrides the CLI command line used for the 'claude' or 'codex' search backends.
TYPESAFE_API_KEYYesRequired. Your TypeSafe API key, used to authenticate with the TypeSafe API.
SIEVE_SEARCH_BACKENDNoOptional. Search backend to use: 'brave', 'claude', or 'codex'. Defaults to 'brave' if BRAVE_API_KEY is set, otherwise 'claude'.
SIEVE_SEARCH_TIMEOUTNoOptional. Timeout in seconds for search backend calls. Defaults to 90.

Instructions

Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.

This server publishes no instructions, or was last inspected before Glama recorded them.

Capabilities

Features and capabilities supported by this server

Protocol revision2025-11-25

CapabilityDetails
tools
{
  "listChanged": false
}
prompts
{
  "listChanged": false
}
resources
{
  "subscribe": false,
  "listChanged": false
}
experimental
{}

Tools

Functions exposed to the LLM to take actions

NameDescription
jev_grepB

Rank the files, or the functions inside the strongest files, of a repository by how relevant they are to a plain-language question. Returns {path, line_start, line_end, kind, probability} sorted by probability, plus token and cost usage. Enumeration is gitignore-aware; answers are cached per (model, question, unit).

jev_rankB

Score each candidate for relevance to a question and return [{id, probability}] sorted by probability, plus token and cost usage. Candidates are truncated to 2000 characters and sent 40 per request.

jev_searchA

Search the web for a plain-language query and return [{url, title, snippet, probability}] sorted by probability. sieve proposes 2-4 query variants in code, runs them concurrently through the configured backend (brave, headless claude, or headless codex), dedupes by canonical url, and reranks everything against your original query with Jev. Snippets are empty on the CLI backends; codex titles are model-transcribed.

Prompts

Interactive templates invoked by user choice

NameDescription

No prompts

Resources

Contextual data attached and managed by the client

NameDescription

No resources

TDQS

A4/5.0

Scored across 3 tools

Disambiguation5/5

Each tool targets a clearly distinct data domain: web search results (jev_search), repository code units (jev_grep), and arbitrary candidate text (jev_rank). Despite all returning probability-ranked output, the input types and descriptions make selection unambiguous.

Naming Consistency5/5

All tools share the jev_ prefix followed by a simple verb: search, grep, rank. This predictable verb-oriented pattern makes the tool names easy to learn and distinguish.

Tool Count5/5

Three tools is minimal but well-scoped for a focused relevance-ranking server. Each tool handles a distinct retrieval or ranking task, and none feels redundant or missing.

Completeness5/5

The set covers the core relevance-ranking workflow: search the web, find relevant code within a repository, and rerank arbitrary candidates. There are no obvious dead ends or critical missing operations for the stated purpose.

Maintenance

ActivityMaintained
ResponsivenessNo issues