glance — LLM-optimized output summarizer

LLMs read every token of command output with equal attention, unlike humans who skim head/tail and scan for patterns. glance gives LLMs a human-like "skim" — pipe output in, get a token-efficient summary with head/tail + regex-matched lines, plus an ID to drill deeper.

Install

# Install permanently (puts `glance` on PATH)
uv tool install glancecli

# One-shot without installing
uvx --from glancecli glance

# Or with Go
go install github.com/juxt/glance@latest

Claude Code plugin:

/plugin marketplace add juxt/glance
/plugin install glance

What it does

$ seq 500 | glance -n 3
  1: 1
  2: 2
  3: 3
498: 498
499: 499
500: 500
--- glance id=20260219-143022-a3f8b1c0 | 500 lines | showing 6 | sections: 1-3, 498-500 ---

$ (seq 100; echo "ERROR db connection refused"; seq 200) | glance -n 3 -p errors
  1: 1
  2: 2
  3: 3
101: ERROR db connection refused
299: 198
300: 199
301: 200
--- glance id=20260219-143025-b7c2e4f1 | 301 lines | showing 7 | sections: 1-3, 101, 299-301 ---

The stored capture can be drilled into:

$ glance show 20260219-143025-b7c2e4f1 -a 101 2
 99: 99
100: 100
101: ERROR db connection refused
102: 1
103: 2
--- glance show 20260219-143025-b7c2e4f1 | 301 lines | showing 5 | sections: 99-103 ---

Design decisions

• Single static binary — compiled Go, no runtime dependencies. Cross-compiled for Linux, macOS, and Windows (amd64 + arm64).
• OR semantics — all matchers (filters + presets) OR together. Head/tail always shown. This is the most useful behavior for scanning output: "show me the start, end, and anything interesting".
• Persistent storage — captures stored in $XDG_CACHE_HOME/glance/captures/ with timestamp + hex IDs (e.g. 20260219-143022-a3f8b1c0). Full ID required for glance show — use glance list to find IDs.
• No metadata files — line count and age derived from the stored file itself (wc -l, stat).
• Built-in + user presets — three hardcoded presets (errors, warnings, status) cover common patterns. User presets stored in ~/.config/glance/presets.csv as CSV. Use (?i) prefix for case-insensitive matching.

Usage

See glance help for complete documentation. Key commands:

Preset names must be alphanumeric (plus hyphens and underscores). Use (?i) prefix in regex for case-insensitive matching:

glance presets add deploys '(?i)deploy|release|rollout' 'Deployment events'

Testing

go test ./...

Files

• main.go — entry point and CLI
• *_test.go — integration and unit tests
• .github/workflows/publish.yml — PyPI release workflow
• README.md — this file