Structured context for AI coding agents and human code review. Evidence-based analysis that powers assistants, pipelines, and teams. Local-first. Deterministic. Fast.
Local-first CLI that analyzes git diff and provides structured context for AI coding agents and human code review. Prompt-ready diffs. Risk scoring. Evidence-based signals. Zero network calls.
Deterministic CLI that extracts high-signal context bundles from repositories for AI-assisted coding. Budget-aware selection. Import graph traversal. TypeScript and Python support.
CLI for persistent AI agent memory through a three-layer technical knowledge graph. Facts, decisions, and agent rules. Versioned, queryable, and corruption-safe.
Structured context for AI coding agents and human code review. Evidence-based output with findings tied to diff evidence.
Prompt-ready diffs with smart filtering. Delta mode for tracking changes. Integrates with Cursor, Claude, Jules, and OpenCode.
Risk scoring with evidence-backed flags. Markdown reports for code review. SARIF output for GitHub Code Scanning.
Deterministic and offline. No LLMs, no network calls. Your code never leaves your machine.
Framework-specific analysis for SvelteKit, Next.js, React, Vue, Astro. Route detection, API changes, and migration safety checks.
Save workspace snapshots for agent iteration. Compare runs with delta mode. Stable IDs for tracking findings across sessions.
Summaries and risk flags tied to diff evidence. File paths, line numbers, and findings that AI and reviewers can verify.
## Summary
This PR modifies 12 files across 3 categories.
### Route Changes
+ Added: /dashboard/settings
+ Added: /api/v2/users
- Removed: /api/legacy/auth
### Breaking Changes
! API endpoint removed: POST /api/legacy/auth
! Environment variable renamed: API_KEY -> AUTH_TOKEN
### Risk Assessment
Score: 67/100 (Medium)
Flags: breaking-api, env-change{
"highlights": [
"Breaking API change: legacy auth endpoint removed",
"New dashboard settings route added",
"Environment variable migration required"
],
"findings": [...],
"riskScore": {
"score": 67,
"level": "medium",
"flags": ["breaking-api", "env-change"]
}
}+------------------------------------------+
| RISK REPORT |
+------------------------------------------+
| Score: 67/100 |
| Level: MEDIUM |
+------------------------------------------+
| Flags: |
| - breaking-api (weight: 20) |
| - env-change (weight: 15) |
| - route-removal (weight: 10) |
+------------------------------------------+Extract high-signal context bundles for AI-assisted coding workflows. Budget-aware selection that fits within LLM context windows.
Start from entry files, symbol searches, git diffs, or error logs. Flexible anchor types for different workflows.
Automatically includes dependencies up to configurable depth. Understands TypeScript and Python import relationships.
Enforces strict character and token limits. Smart ranking ensures highest-signal code is included first within your budget.
Same inputs always produce identical bundles. Important for reproducibility in CI/CD and automation pipelines.
Auto-detects npm and Python workspaces. Scope queries to specific packages or search across the entire monorepo.
Parse TypeScript, Jest, pytest, mypy, and pyright errors. Automatically bundle code relevant to build failures.
# Context Bundle
# Budget: 12,450 / 28,000 chars
## Index
1. src/auth.ts (definition, score: 1000)
2. src/middleware.ts (import-depth-1, score: 800)
3. src/types/user.ts (import-depth-1, score: 750)
4. src/utils/token.ts (import-depth-2, score: 600)
## src/auth.ts
export function handleAuth(req: Request) {
const token = extractToken(req);
return validateUser(token);
}{
"metadata": {
"budget": { "chars": 28000, "used": 12450 },
"anchors": ["symbol:handleAuth"],
"depth": 2
},
"items": [
{
"path": "src/auth.ts",
"reason": "definition",
"score": 1000,
"content": "..."
}
]
}# Bundle from git diff
$ repo-slice pack --from-diff origin/main..HEAD
# Analyzing changed files...
+ Found 5 changed files
+ Expanding import graph (depth: 2)
+ Included 12 related files
+ Bundle: 18,200 / 28,000 chars
# Perfect for PR review contextPersistent memory for AI agents through a three-layer knowledge graph. Facts, decisions, and operating rules that survive across sessions.
Entity-centric facts stored in versioned, queryable format. Organized by projects, libraries, patterns, and conventions.
Decision rationale and trade-offs documented in DECISIONS.md. Capture why choices were made, not just what was chosen.
Operating principles and constraints for AI assistants in AGENTS.md. Generate config for Claude, Cursor, or Codex.
Queue insights during active work sessions without interrupting flow. Review and commit facts when ready.
Search facts by category, entity, or keyword. View complete history including superseded entries and version chains.
Export and import knowledge graphs. Atomic file operations prevent corruption during concurrent access.
# Extract facts from conversations
$ memo extract "Next.js 15 uses turbopack by default"
+ Entity: next.js
+ Category: libraries
+ Fact: "Uses turbopack by default in v15"
+ Confidence: high
# Queue insights during work
$ memo draft "Zod v4 drops .parse() in favor of .check()"
+ Queued to drafts (3 pending)# Query knowledge by entity
$ memo query --entity react
## react (6 facts)
1. Uses compiler for automatic memoization (v19)
2. Server Components are the default (v19)
3. use() hook replaces useEffect for data fetching
4. Actions replace form event handlers
5. Ref forwarding no longer needs forwardRef
6. Suspense works with SSR streaming
# Search across all entities
$ memo query --search "compiler"
+ Found 3 facts across 2 entities# Generate AI agent config
$ memo generate --target claude
+ Generated CLAUDE.md
+ Included 42 facts from knowledge graph
+ Included 8 technical decisions
+ Included 5 agent rules
# Synthesize summary
$ memo synthesize
+ Generated summary from 42 current facts
+ Covering 12 entities across 4 categoriesBetter Vibe builds developer tools that provide structured context for AI coding agents and human code review. We prioritize evidence over speculation, determinism over magic, and privacy over cloud dependencies.
Our tools bridge the gap between AI agents and human reviewers. Structured output that agents can consume, evidence that developers can verify.
Install our tools to give AI agents structured context and human reviewers evidence-backed insights.
npx @better-vibe/branch-narrator integrate
npx @better-vibe/repo-slice pack --symbol "main"
npx @better-vibe/memo init --agent