Built for flow state
A full VS Code-class editor with AI wired into every surface — and a local-first architecture that keeps your code on your machine.
💬 EditVolt Chat
Five modes, one panel — from questions to autonomy
Switch between Ask, Plan, Debug, Agent, and Multitask — from read-only Q&A to parallel autonomous workers. Ask and Plan never touch your files until you approve.
- @-mention context: files, folders, symbols, problems, git changes, terminals, docs, and URLs
- Attach images — vision works across all providers
- Debug mode gathers evidence before it acts
- Multitask runs several workers in parallel
You: @Workspace @Problems why is the build failing?
EditVolt: The failure comes from
src/api.ts:118 — a type error introduced by
the change in @Changes. Switch to Agent mode
and I'll fix it.
⚡ EditVolt Complete
Completions that work at 30,000 feet — literally
Inline completions run on a bundled local engine — no cloud round-trip, no API key, works fully offline. Cloud completions are available as an explicit opt-in.
- Streaming multi-line suggestions as you type
- Tab-to-jump: predicts your next edit and takes you there
- Change propagation across related edits
- Context from nearby code and recent edits
function formatPrice(cents) {
return new Intl.NumberFormat("en-US", {
style: "currency", currency: "USD",
}).format(cents / 100); Tab
}
⚡ local engine · offline · 0 tokens sent
🤖 Agent Mode
Autonomy with a paper trail
The agent reads, edits, and runs commands with a full tool surface — but every edit is staged as a diff for your approval, never auto-applied. Checkpoints let you roll the workspace back to any point in the conversation.
- "Review My Changes" unified multi-file diff
- Command consent levels: auto-review, allowlist, or run-everything
- File-deletion protection on by default
- Agents window: start a project from an empty window with built-in scaffolding
⚡ Task: add rate limiting to the API
✓ Plan created (3 steps)
✓ Staged: src/middleware/rateLimit.ts +64
✓ Staged: src/api.ts +3 −1
✓ Tests passed (24/24) · Review diff · ⟲ Checkpoint
🧠 Local Codebase Intelligence
Your repo, semantically searchable — on your machine
EditVolt builds a vector index of your codebase using on-device embeddings — the embedding model ships with the app and never touches the network. Semantic search powers @Workspace answers and the agent's code discovery.
- On-disk vector database, per workspace
- Bundled embedding model — indexing works offline
- Hybrid retrieval: semantic + fast text search
- Optional local memory that learns your project (off by default)
⚡ Workspace index
✓ 1,284 files indexed · local embeddings
✓ 0 bytes sent to the network
✓ 14 paths excluded by ignore file
semantic_search("webhook signature validation")
→ src/webhooks/verify.ts:42 (0.94)
🔌 Bring Any Model
Direct to the provider — no middleman, no markup
Connect the models you already pay for. Requests go straight from your machine to the provider — EditVolt has no proxy in between, and model lists update live from each provider.
- Anthropic Claude — API key, or bridge your existing Claude Code CLI sign-in
- Google Gemini with your AI Studio key
- Any OpenAI-compatible endpoint: OpenAI, OpenRouter, Azure, vLLM, LM Studio
- Hugging Face — remote or local
- Ollama for 100% local, key-free inference
Models
● Claude — connected (your key)
● Gemini — connected (your key)
● Ollama — local · 3 models
○ OpenAI-compatible — add endpoint…
Routing: fast lane → local · heavy → Claude
🧩 Make It Yours
Rules, skills, subagents, hooks — a real customization substrate
Shape how the AI works in your repo with plain files in the project config folder — versionable, reviewable, shareable with your team.
- Rules packs injected into every conversation
- Skills and subagents for repeatable workflows (security review, docs editing…)
- Slash commands: /explain, /fix, /review, /test, /architect, /why, /scaffold
- Pre/post tool-use hooks and JSON automations
- MCP servers with per-tool consent and allowlists
- One-click import of your Cursor, Claude, or Copilot configs and AGENTS.md
.editvolt/
├─ rules/ no-hardcoded-secrets.md
├─ skills/ split-to-prs.md
├─ subagents/ security-reviewer.md
├─ hooks/ pre-commit-check.sh
├─ commands/ /review · /scaffold
└─ permissions.json
🔒 Privacy & Security
Not a promise — an architecture
Privacy isn't a settings toggle in EditVolt; it's how the product is built. The telemetry pipeline is structurally removed from the binary, and an audit surface in the app shows you the outbound-connection count in real time.
- Zero telemetry, zero crash uploads, zero experiments — verified by network audit
- Strict privacy mode: one setting forces all AI to local models
.editvoltignore keeps secrets out of AI context — on by default for .env files, keys, and credentials- API keys live in your OS keychain, read per-request
- Extensions from Open VSX — no Microsoft marketplace, no Microsoft telemetry
- Cert-pinned auto-updates and network-audited releases
⚡ Audit
✓ telemetry-zero · structurally removed
✓ 0 outbound connections this session
✓ ignore file active · 14 paths fenced
✓ keys: OS keychain · never on disk