Skip to content

Home

Jump to bottom
github-actions[bot] edited this page Mar 15, 2026 · 2 revisions

ContextBudget Wiki

ContextBudget selects, compresses, and budgets repository context for coding-agent workflows. It is deterministic, local-first, and produces stable machine-readable artifacts for reuse in CI, local tooling, and agent middleware.

Quick Start

pip install -e .[dev]

# Rank relevant files
contextbudget plan "add caching to search API" --repo .

# Pack context under budget
contextbudget pack "add caching to search API" --repo . --max-tokens 30000

# Summarize the run artifact
contextbudget report run.json

Key Features

Feature Description
File Ranking Deterministic relevance scoring against natural-language tasks
Token Packing Packs context under explicit token budgets with quality-risk estimation
Compression Snippet extraction, symbol extraction, language-aware chunking, and deterministic summaries
Incremental / Delta Re-packs only changed files across agent loop iterations
Workspace Support Multi-repo and monorepo-package scanning with repo provenance tracking
Benchmarking Compares full-context, top-k, compressed, and cache-assisted strategies
Policy Enforcement Strict-mode CI gates on token count, file count, and quality risk
Agent Middleware Adapter-ready middleware layer for external agent tools
Stable Artifacts run.json and run.md with additive metadata blocks

Pages

Page Description
Getting Started Installation and core workflow
CLI Reference All CLI commands and flags
Python API BudgetGuard and ContextBudgetEngine reference
Configuration contextbudget.toml schema and model profiles
Architecture System layers, pipeline stages, and design goals
Agent Integration Middleware, adapters, and multi-turn agent loops
Plugins Custom scorers, compressors, and token estimators
Benchmarking and Diff Strategy comparison and run diffing
Policy and CI Budget policy enforcement and GitHub Actions
Workspace Multi-repo and monorepo workspace configuration

Requirements

  • Python 3.11+
  • Install: pip install -e .[dev]
  • Optional exact tokenization: pip install -e .[tokenizers]

Clone this wiki locally