Build your first Flame workflow
A practical guide to the Flame IDE: open your workspace, move through code, run terminals, delegate to agents, and land work without losing control.
From install to useful in minutes
Start with one workspace root. Flame will discover projects from there and keep the rest of the app organized around that root.
Install Flame
Download the desktop build, launch it, and let Flame start its local workspace server. The app keeps that server on loopback unless you enable network access.
Choose a projects root
Pick the folder where your repos live. Flame scans it for projects and shows branch, change count, package manager, project type, and active worktrees.
Open a project
Use the project switcher or command palette to jump into a repo. Flame restores your recent tabs and can reopen terminals from the last session.
Add AI credentials
Open Settings, then AI and Model providers. Add Claude, OpenAI, Gemini, an OpenAI-compatible endpoint, Ollama, or LM Studio depending on how you want agents to run.
The pieces Flame is built around
Once these words click, the rest of the IDE feels straightforward.
Workspace
The top-level environment Flame watches: projects, groups, sessions, terminals, agents, settings, and file history.
Project
A repo or app inside your projects root. Each project gets its own tabs, terminals, git state, scripts, diagnostics, and agent context.
Group
A named set of projects. Use groups to filter the sidebar, broadcast commands, and apply shared agent or merge policies.
Worktree
A second checkout of the same repo on another branch. Flame uses worktrees so you and your agents can work in parallel without stashing.
Agent job
A managed AI task with a prompt, model, permissions, status, logs, optional schedule, and a review path before it lands.
Merge queue
The review and landing lane for completed agent work. It can gate on commands, handle conflicts, and merge, push, or open a pull request.
Common workflows
Use these as the first pass through Flame. Each one maps to controls that already exist in the desktop app.
Open your daily stack
- Set the projects root to the parent folder that contains your repos.
- Pin the projects you touch most often, then create groups for stacks or clients.
- Use smart file filters such as Source, Tests, Config, Env, Components, Routes, API, CI/CD, and TODO/FIXME to narrow large trees.
- Enable session restore if you want Flame to reopen tabs and terminals when you come back.
Move through code quickly
- Use quick open for files and the outline or breadcrumbs for symbols.
- Open side-by-side diffs from git, file history, pasted content, or another branch.
- Turn on Vim mode, Emmet, format on save, and format on paste from editor settings if they match your muscle memory.
- Preview images, PDFs, notebooks, CSV/Excel files, archives, fonts, audio, and video without leaving the workspace.
Delegate a task safely
- Open AI Agents Manager, write the task, then choose the project, model, branch, and landing policy.
- Use isolation when the agent should work in a fresh git worktree under the workspace .worktrees folder.
- Watch tool calls, terminal output, token use, and status live. Stop, duplicate, retry, or requeue the job at any point.
- Review the diff when it finishes, then commit, merge, push, open a pull request, or discard the worktree.
Keep branches calm
- Create worktrees from the git panel when you need two branches open side by side.
- Stage files or hunks from Changes, inspect inline blame and file history, and use the commit graph to understand the branch.
- Configure project merge gates such as tests or builds before agent work is allowed to land.
- If a rebase or merge conflicts, ask the agent to resolve in its own worktree, then rerun the gate commands before landing.
Run commands where they belong
- Open integrated terminals per project or pop one out to a separate window.
- Run package scripts from the script runner when Flame detects package.json.
- Use broadcast terminal to run the same command across a group, every project, or selected worktrees.
- Track dev server ports and open previews from the workspace instead of hunting through terminal output.
Pick the right AI provider
- Use Claude through login, API key, or token when you want the bundled agent runtime.
- Use OpenAI, Gemini, or a custom OpenAI-compatible base URL for chat, completion, and one-shot AI actions.
- Use Ollama or LM Studio when you want local model calls and offline experiments.
- Set defaults per workflow, then switch model or provider on individual conversations when the task calls for it.
The details you check twice
A compact reference for languages, model permissions, and the safety defaults that shape daily work.
Language intelligence
- Bundled
- TypeScript, JavaScript, HTML, CSS, JSON, YAML, Bash, Dockerfile, PHP, Vue, Svelte, Astro, Markdown.
- Native when present
- Python, Rust, Go, C/C++, Java, C#, Kotlin, Swift, Dart, PowerShell.
- Linters and formatters
- ESLint, Biome, Stylelint, Ruff, Prettier, and Biome formatting.
Agent permissions
- Scopes
- Global, group, project, or current chat.
- Actions
- File reads and writes, terminal and git, browser control, IDE control.
- Decisions
- Allow once, deny, save a rule, or keep asking.
Security defaults
- Network
- Loopback by default. Same-Wi-Fi access is opt-in and may require relaunching Flame.
- Credentials
- Tokens and provider credentials are stored by the desktop app and encrypted at rest where the platform supports it.
- Review
- Agent edits, commands, and landing actions stay visible through tool cards, logs, diffs, and the merge queue.
Daily shortcuts
Flame is meant to stay under your hands. Start with these, then open the cheat sheet for the full map.
Keep going from here
The docs page is the short path. These pages go deeper into the surfaces you will use most.
Ready to try it yourself?
The fastest way to learn Flame is to open a real repo, run a real terminal, and hand one small task to an agent.
No spam, ever. Just your invite and the occasional release note.