Questions, answered

Flame is an AI-native IDE: a full development workspace for people who juggle many projects at once. It pairs the editor you know with multi-project workspaces, git worktrees, an integrated and broadcast terminal, live previews, and an Agents Manager for AI tasks.

Flame features a powerful editor that feels immediately familiar, with all the standard shortcuts and features. What's different is everything around it: a real multi-project workspace, git worktrees so branches live side by side, a broadcast terminal that runs one command across many repos, a cross-project dependency diagram, and an Agents Manager, not a chat box bolted onto a single folder.

Bundled tooling covers TypeScript/JavaScript, HTML, CSS/SCSS/LESS, JSON, YAML, Bash, Dockerfile, PHP, Vue, Svelte, Astro, and Markdown. Flame also drives language servers for Python, Rust, Go, C/C++, Java, C#, Kotlin, Swift, Dart, and PowerShell when they are available on PATH. Linters attach when the required tool or project configuration is present.

Desktop builds target macOS on Apple Silicon and Intel, Windows x64 and ARM64 installers, and Linux as an x64 or arm64 AppImage plus an x64 .deb package.

Yes. Flame supports breakpoints, stepping, variables, watches, and call stacks over the Debug Adapter Protocol. When you use Flame's built-in Node debug engine, the profiler can record interactive CPU flame graphs, mark hot source lines, capture heap snapshots, and compare allocations.

Yes. Flame can install and manage local manifest extensions with declarative contributions for languages, snippets, themes, formatters, and linters. You can also customize themes, density, shortcuts, gestures, snippets, formatters, and project or group run configurations, then move portable configuration between installations as scrubbed JSON.

The Test Explorer has built-in adapters for Vitest, Jest, Node's test runner, Pytest, Go test, and Cargo test. It supports discovery, run and watch workflows, grouped results, and coverage gutters where coverage data is available.

Flame runs Claude Code, OpenAI Codex, and Google Gemini through the Antigravity CLI as full coding agents, using your existing Claude, ChatGPT, or Google sign-in. You can also connect Anthropic, OpenAI, Gemini, OpenRouter, Bedrock, Vertex, Ollama, LM Studio, vLLM, LiteLLM, or another compatible endpoint for chat, inline completion, and focused AI actions, with defaults per surface and model overrides per conversation.

Yes. Each task can run in its own git worktree and branch, so multiple agents work in parallel without touching your working tree, and you set a concurrency limit for the whole fleet.

You choose each job's project, permissions, branch strategy, and landing policy. A task can run in an isolated worktree, while tool activity, diffs, test results, and conflicts stay visible. Manual review is the default; automatic landing is opt-in.

Plan Mode locks the agent into a strict, server-enforced read-only sandbox where it can inspect files, search symbols, and read diagnostics, but cannot edit files or run mutating shell/database commands. It generates a structured Markdown plan that opens in an editable editor tab. You can review, refine, or request changes with conversational feedback; write tools are unlocked only when you explicitly approve the plan.

The core workspace is local-first and works without a cloud service. For AI, use Ollama, LM Studio, vLLM, or another local compatible endpoint to keep model requests on your machine; update checks and other integrations connect only when you use them.

By default, your code stays local. Flame's server binds to 127.0.0.1 and reads your files on your device. A connected AI or developer service receives only the context needed for an action you choose. With a local model, AI requests stay on your machine.

Encrypted at rest on your machine (protected by the OS keychain in packaged builds) and never echoed back to the UI. Integrations like GitHub/GitLab tokens are stored the same way.

Yes. Security-relevant actions (opening a terminal, writing a file, pushing to a remote, and every tool an agent invokes) are recorded to an audit log that distinguishes you from the agent.

Only if you turn it on. Flame can share your version and operating system — six values like "0.1.9" and "Windows 11", plus a random number generated on your computer — to help us decide which versions and platforms to support. It is off by default, you can see the exact message before deciding, and turning it off deletes the random number. It never carries your name, files, project names, paths, or anything you type. Crash reports are separate and still manual-send: written locally, reviewable in full, transmitted only if you choose, with home-directory and OS username details redacted first.

Anywhere mode keeps the Flame server on loopback while the host dials out through an end-to-end encrypted, content-blind relay. The relay cannot decrypt workspace traffic, but it can observe limited connection metadata such as timing, byte volume, and live socket IP addresses. Pairing requires a short-lived QR code, a matching six-word host fingerprint, and a two-digit code entered on the host. The browser companion delivered from connect.flame-ide.comremains part of Flame's trust boundary.

Flame is available directly on our download page for macOS, Windows, and Linux. Choose your operating system and start building.

Flame is free to use.