Agents Manager
Create, monitor, stop, retry, schedule, and review AI tasks across projects, with a workspace-wide concurrency limit.
Run your own shell commands on agent lifecycle events such as prompt submitted, tool use, waiting for you, and validation failed, at global, group, or project scope.
Native delegated workers
In supported Claude sessions, enable Multi-agent mode for read-only research and review workers while the lead owns edits and commands. Inspect recorded Claude and Copilot worker activity; support depends on runtime capabilities.
Chat history transfer
Import supported Claude Code, Codex, Copilot, Claude and ChatGPT exports, or a Flame chat archive. Preview and select histories, map them to projects, and export selected conversations without changing source histories.
AI chat in context
Work with selected files, diffs, terminal output, test results, and browser state under scoped, revocable permissions.
Conversation forks
Fork from a message into an independent chat or editor tab, navigate its parent and related forks, and retain the original conversation. Forking keeps checkout context; choose a worktree separately for file isolation.
Usage-limit continuation
Schedule a usage-limited turn to continue when access resets or at a chosen local time. Pending schedules persist across restarts; the backend must be running, and uncertain in-flight sends pause for review.
Multiple & global chats
Keep several conversations open as tabs, start a project-free chat when appropriate, search models, rename or auto-title chats, and clear or archive the context you no longer need.
Choose where each chat works
Run a chat or a task on the current branch, a new branch, any other branch, or its own Git worktree, and move a running chat between checkouts or projects.
Send later or after all chats
Queue or steer while the agent works, schedule a message for a date and time, or send it once every chat in the project is idle.
Images & prompt polish
Paste or drop images into chat, attach curated project context, and improve a prompt with reviewable grammar, clarity, and rewrite suggestions before sending.
Choose your agent
Run Claude Code, OpenAI Codex, or Gemini through Google Antigravity as a full coding agent, and use GitHub Copilot in AI Chat. Switch models per conversation.
Bring your model
Use a local model through Ollama or LM Studio, or connect Anthropic, OpenAI, Gemini, OpenRouter, Bedrock, Vertex, or another compatible endpoint.
Local voice control
Use local Whisper transcription to dictate prompts and trigger supported navigation or agent intents.
Inline completion & next-edit
Fill-in-the-middle completions and Cursor-style next-edit predictions, grounded in your project's symbols.
Focused AI actions
Hand off a failing test, terminal error, API response, diff, review finding, merge conflict, profiler hotspot, or project configuration from the surface where it appears.
Models per assistant
Pick the provider and model for Generate commit message, Summary with AI, Improve writing, Is this safe?, and Help me decide separately from chat.
Queued & dependent work
Schedule an agent for a specific time, wait for another job or an idle conversation, and chain dependent tasks while retaining each job's status, token, duration, and tool history.
Internal browser-to-agent handoff
Send a page, selected elements, screenshots, DOM state, and console messages from Flame's internal browser straight into an AI conversation.
IDE control with confirmation
Let an agent navigate and invoke registered IDE commands through a risk-tiered bridge that confirms mutating or sensitive actions.
Interactive Plan Mode
Enforce a server-owned planning and review lifecycle (planning → review → executing → completed). Read-only tool gates let agents prepare a plan you review in the Plan tab, with decisions, recommendations, and an architecture diagram, before you approve execution.
Structured questions
When an agent needs a decision mid-run, it can ask a structured question and continue from your answer.