Skip to content

DevDiff — Known Limitations & Scope

A transparent and factual guide to DevDiff's operational boundaries, capabilities, performance expectations, and versioning lifecycle.


1. AI Reasoning & Inference Boundaries

LimitationTechnical ReasonRecommended Solution / Workaround
Massive Diffs (>500 files or >10k LOC)Exceeds LLM context window limits ($8\text{k}-128\text{k}$ tokens).DevDiff automatically applies hierarchical AST compression. For line-by-line detail, stage smaller commits.
First-Run Indexing on Large ReposAST indexing parses all workspace symbols on cold start.Run devdiff memory init or let the background indexer complete its initial scan ($30-60\text{s}$ for $100\text{k}+$ files).
Auto-Generated Code & LockfilesHigh noise ratio with little semantic architectural value.Ignore generated assets via .devdiffignore (e.g. pnpm-lock.yaml, dist/, build/).
Internal Domain JargonGeneral LLMs lack proprietary company acronyms.Define project-specific terms and architecture in .devdiff/SKILL.md.
Hardware-Dependent Local InferenceLocal models run on host GPU/CPU hardware.On machines without dedicated GPU VRAM, use quantized models (e.g. llama3.2:3b) or lightweight cloud providers.
Binary & Minified FilesCompiled binaries and minified bundles lack readable diff lines.DevDiff reports binary and minified file modifications by status and byte size without generating natural prose.

2. Platform & Environment Compatibility

Platform / EnvironmentStatusNotes
macOS (Apple Silicon & Intel)✅ Native SupportHardware acceleration supported via Metal on Apple Silicon.
Windows 10 / 11 & Windows Server✅ Native SupportFull support in PowerShell, CMD, and Windows Terminal.
Linux (x86_64 & ARM64)✅ Native SupportFully compatible with headless servers, CI runners, and devcontainers.
WSL2 (Windows Subsystem for Linux)✅ SupportedAutomatically detects host Ollama daemons across localhost bridging.
Docker & Air-Gapped Containers✅ Supported100% offline execution with local Ollama, ONNX Transformers.js, or template fallbacks.
CI/CD Pipelines (GitHub Actions, GitLab CI)✅ SupportedAutomated headless execution via devdiff --ci or pnpm gate.

3. Strict Privacy & Execution Guarantees

DevDiff operates under a strict, human-craft privacy and safety contract:

  • No Silent Network Traffic: Outbound network connections are blocked by default unless explicitly configured for a cloud AI provider.
  • No Background Code Modification: DevDiff never modifies, commits, or pushes source code automatically.
  • No Telemetry or Tracking: Zero tracking, analytics, fingerprinting, or user behavior collection.
  • No Account Requirements: No sign-ups, API keys, or credit cards needed for local AI execution.
  • No Model Training: Your codebase diffs and context are never transmitted to training corpuses.

4. Performance Expectations

Changed FilesTypical Analysis LatencyMemory FootprintFallback Strategy
1 – 20 files$<1.5\text{ seconds}$$<35\text{ MB}$Direct line-by-line synthesis
20 – 100 files$1.5 – 5\text{ seconds}$$<50\text{ MB}$Module-level AST aggregation
100 – 500 files$5 – 15\text{ seconds}$$<75\text{ MB}$Architectural subsystem grouping
>500 files$15 – 30\text{ seconds}$$<100\text{ MB}$Deterministic structural summary fallback

5. Versioning Lifecycle & Strict SemVer 2.0.0

DevDiff follows strict Semantic Versioning 2.0.0:

$$\mathbf{\text{MAJOR}}.\mathbf{\text{MINOR}}.\mathbf{\text{PATCH}}$$

Current Release: v1.9.0

The current release milestone across the monorepo, npm, VS Code Marketplace (ebula.devdiff), and Open VSX Registry (ebula/devdiff) is v1.9.0.

Forward Release Roadmap:

  • v1.9.1 (Next Patch): Reserved for backward-compatible bug fixes, performance optimizations, and security patches.
  • v1.10.0 (Next Minor): Targets upcoming feature additions, new CLI/MCP tool integrations, and expanded provider drivers.
  • v2.0.0 (Future Major): Reserved for breaking API contracts, architectural changes, or configuration schema evolutions.

6. Reporting Issues & Community Support