Textor Nexuum ??

Weave AI clarity into every GitHub issue.

Textor Nexuum is a production-minded Python CLI named nexuum. It triages GitHub issues with Grok, OpenAI-compatible APIs, and now Anthropic Claude via the same structured JSON contract. Output stays safe by default (dry-run first; writes only with --apply).

• Repository: FratresMedAI/TextorNexuum
• PyPI: textor-nexuum
• Python: 3.11+
• License: MIT

Install

python -m pip install textor-nexuum

From source:

git clone https://github.com/FratresMedAI/TextorNexuum.git
cd TextorNexuum
python -m pip install -e ".[dev]"

Quick start

nexuum --version
nexuum analyze octocat/Hello-World --limit 3 --verbose
nexuum config init

Without GITHUB_TOKEN, built-in mock issues are used so the CLI still demos cleanly.

New in v1.0 line

• Anthropic support: if base_url contains anthropic, Nexuum uses /messages + x-api-key + anthropic-version.
• Configurable duplicate threshold: duplicate_threshold in .nexuum.yaml (default 0.82).
• Better duplicate detection: weighted score on title + first 200 chars of body.
• Verbose mode: --verbose prints issue-level reasoning and token usage.
• API resilience: retry/backoff (3 attempts) for LLM calls with explicit fallback message.

Environment variables

Commands

nexuum analyze owner/repo --limit 10
nexuum analyze owner/repo --issue 42
nexuum analyze owner/repo --format json
nexuum analyze owner/repo --verbose
nexuum analyze owner/repo --apply --yes
nexuum config init
nexuum version

Typer completion is enabled. Install completion with your shell profile if desired (e.g. --install-completion).

Configuration

Resolution order:

1. .nexuum.yaml
2. .weaver.yaml (legacy)
3. ~/.config/nexuum/config.yaml

Example .nexuum.yaml:

model: claude-3-5-sonnet-latest
base_url: https://api.anthropic.com/v1
max_issues: 30
duplicate_threshold: 0.82
allowed_labels: [bug, feature, documentation, question, enhancement]
custom_labels: [needs-triage, good-first-issue]
prompt_additions: ""

Real-world examples

Grok / OpenAI-compatible

nexuum analyze owner/repo --limit 15 --model grok-4.1-fast --base-url https://api.x.ai/v1

Anthropic Claude

set ANTHROPIC_API_KEY=your-key
nexuum analyze owner/repo --limit 15 --model claude-3-5-sonnet-latest --base-url https://api.anthropic.com/v1

Single issue with detailed diagnostics

nexuum analyze owner/repo --issue 123 --verbose

How it works

1. Load config + CLI flags into NexuumConfig.
2. Fetch issues from GitHub (or mock fixtures when unauthenticated).
3. Build prompt from src/nexuum/prompts.py.
4. Call selected LLM provider (OpenAI-compatible or Anthropic).
5. Parse strict JSON into TriageResult; fallback to heuristics on failures.
6. Run duplicate checks with configurable threshold and confidence.
7. Render Rich table + exports (md, json); optional guarded apply flow.

Comparison to other tools

• GitHub native triage: Great for manual workflows, but no deterministic structured AI output or local automation.
• Dosu: Strong hosted assistant model, less local CLI control over provider routing/fallback behavior.
• PR-Agent: PR-focused automation; Textor Nexuum is issue-triage focused with explicit duplicate heuristics.
• Sweep: Strong coding automation path; Nexuum is intentionally lightweight triage-first with safe dry-run defaults.

GitHub Action (summary-only)

Use .github/workflows/nexuum.yml: analyze + post one summary comment. It does not run --apply.

Quality gate

ruff check . --fix && ruff format .
mypy src/nexuum --strict
pytest tests/ -q

Support

Issues and PRs: github.com/FratresMedAI/TextorNexuum/issues