🔥 infernoflow — AI Session Memory

Every new AI session starts cold — the gotchas you found, the decisions you made, don't survive. infernoflow makes them stick.

Persistent memory for AI coding sessions. Captures what agents can't infer from code — gotchas, decisions, dead ends — and replays them into your next AI chat (Copilot, Cursor, Claude, Windsurf) so the same mistake never happens twice.

This extension is the visual surface over AMP — the AI Memory Protocol. Memory lives in .ai-memory/ as plain JSONL — per branch, tracked in git so your team shares it — vendor-neutral and file-based.

Features

🤖 Copilot Chat captures memory for you — no setup

infernoflow registers amp_write and amp_read as native VS Code Language Model Tools — so Copilot can recall a past decision on its own and offer to log a gotcha mid-conversation, the moment it notices something worth remembering. Saving asks you first (memory is shared with your team through git). Nothing to wire up: install the extension and Copilot's tool picker shows 🔥 amp_write. You can also call them by hand with #amp_write / #amp_read in the chat box.

These work without any MCP configuration. The CLI also registers its MCP server in .vscode/mcp.json, so Copilot's agent mode can use the full set of amp_* tools; Cursor and Claude Code get them the same way.

A dedicated sidebar shows your session at a glance — health score, gotchas with file paths, decisions in effect, failed attempts to avoid. Click any entry to jump to the file:line where it was logged.

⚠️ Gotchas appear as Problem-panel warnings

Logged a gotcha with a file path? Open that file — the gotcha now shows as a yellow squiggle in the editor and a row in the Problems panel, right next to your TypeScript errors. Both you and Copilot see the warning before making the same mistake again.

📋 One-click handoff for the next AI session

Click 📋 Switch in the status bar or hit Ctrl+Alt+S (Cmd+Alt+S on Mac). The CLI generates a clean gotchas-first markdown handoff and copies it to your clipboard. Paste into any AI chat — the AI now knows about all your project's landmines.

⌨️ Keyboard-first logging

Shortcut	Command
`Ctrl+Alt+G` (`Cmd+Alt+G` Mac)	Log a gotcha
`Ctrl+Alt+D`	Log a decision
`Ctrl+Alt+A`	Ask memory (search)
`Ctrl+Alt+S`	Generate handoff
`Ctrl+Alt+R`	Show session recap

The extension also adds a right-click menu to the editor — log a gotcha for the current line without leaving the file.

🔍 Search your memory

Ctrl+Alt+A opens an inline search across every entry — gotchas, decisions, attempts. Pick a result to jump to the file.

📊 Live status-bar health score

Always visible: 🔥 B 65 · ⚠3 · ✓2 · ❌1 · 📋 Switch. Click the score to open the sidebar; click Switch to copy the handoff.

Requirements

VS Code 1.95+ (the Language Model Tools API that powers Copilot auto-capture is stable from 1.95).
Nothing else for Copilot. The sidebar and Copilot's amp_write / amp_read tools run entirely inside the extension — no Node, no CLI.
For Cursor / Claude Code auto-capture, install the infernoflow CLI (npm install -g infernoflow) so the MCP server those tools speak to gets wired up. The extension offers to install and keep the CLI current for you on first activation.

The extension activates automatically as soon as your workspace contains the AMP layout (or the legacy inferno/ folder). infernoflow init (or logging once from the sidebar) creates .ai-memory/.

Configuration

Setting	Default	What it does
`infernoflow.cliPath`	`infernoflow`	Path to the CLI. Machine-level only (a workspace can't change it). Override if not on PATH.
`infernoflow.showStatusBar`	`true`	Show session-health summary in the status bar.
`infernoflow.showDiagnostics`	`true`	Show gotchas as warnings in the Problems panel.
`infernoflow.notifications`	`important`	`all` / `important` / `none`. Controls confirmation toasts.

How it pairs with the CLI

The extension reads and writes the same files as the CLI, in-process: the shared branch file (.ai-memory/branches/<branch>.jsonl, tracked in git), your personal global.jsonl and this machine's local copy. Entries get the same shape, ids, author stamp and secret redaction as infernoflow log.

Security: in an untrusted workspace (Restricted Mode) memory is read-only and the CLI is never run. Memory shown to the AI is framed as information to verify, not instructions. Requires CLI 0.46.2 or later for the matching behaviour on the CLI side.

switch and recap shell out to the infernoflow CLI for the heavy markdown generation. Everything else is in-process and instant.

License

MIT.

infernoflow