infernoflow
Advanced tools
Persistent memory for AI coding sessions — captures what agents can't infer from code alone. Works with Copilot, Cursor, Claude, and Windsurf.
Persistent memory for AI coding sessions. Capture what agents can't infer from code — the gotchas, the decisions, the failed approaches — and replay it into your next chat so you stop re-deriving context every time you open Cursor / Claude Code / Copilot.
Every new AI session today starts cold. The agent re-reads your code, re-derives the obvious, and often re-makes the same wrong move someone else made yesterday. infernoflow closes that loop in four stages:
amp_write MCP tool — no copy/paste, no git commit -m.gotcha | decision | attempt | note | detection | pattern) with timestamp, file:line, tags, and a stable AMP id. Linked into the project, not your scratchpad..ai-memory/branches/<branch>.jsonl (git-tracked, travels with your branch — teammates inherit it) plus .ai-memory/global.jsonl (personal preferences, gitignored, synced across your machines via any OS-synced folder).CLAUDE.md / .cursorrules / copilot-instructions.md at boot. The most relevant entries are already there. Warm start; no cold derivation.That's it. No service to log into. No SaaS. JSONL on disk, an MCP server, and three rule files your IDE already reads.
npm install -g infernoflow
infernoflow init --yes
Zero runtime dependencies. Works on Node ≥ 18 — macOS, Linux, Windows.
init --yes does the whole setup: creates .ai-memory/, writes rule files for every supported IDE, wires the MCP server for Cursor / VS Code Copilot / Claude Code in one shot, applies the clean-tree git policy, and drops a visible demo entry so you can confirm the loop is alive with one command:
infernoflow status
These cover 95% of usage:
| Command | What it does |
|---|---|
infernoflow log "..." | Remember a gotcha / decision / attempt / note. --type gotcha|decision|attempt|preference |
infernoflow ask "..." | Search your memory by keyword — gotchas surface first |
infernoflow switch | Generate a handoff for the next session. --copy puts it on your clipboard |
infernoflow recap | End-of-session summary with health score + unlogged-change detection |
infernoflow status | Quick health check — entries, gotchas, decisions, last activity |
infernoflow refresh | Manually rebuild CLAUDE.md / .cursorrules / copilot-instructions.md from memory |
In practice you barely run any of these — the MCP-aware AI does it for you. The CLI is for grep-style introspection.
infernoflow commands shows the full list (17 commands, grouped by purpose).
Your teammate takes your branch — they inherit your memory.
.ai-memory/
├── branches/
│ ├── main.jsonl ← project-wide truths (git-tracked)
│ └── feature-auth.jsonl ← your current branch's work (git-tracked)
├── global.jsonl ← your personal preferences (gitignored)
└── sessions.jsonl ← legacy flat file (still read)
git checkout feature-auth, the JSONL is there. Their MCP server boots, reads it, regenerates their rule files — their AI is warm-started on your findings without you sending a message.infernoflow sync set ~/Dropbox/infernoflow-memory
merge=union on branch JSONLs means concurrent commits from different machines merge cleanly — no manual conflict resolution.infernoflow refresh, not on every entry — your working tree stays clean while you log.The rule files at the top level of your project are what every AI agent reads on boot. infernoflow writes the same canonical block to all three:
| Tool | Reads from |
|---|---|
| Claude Code | CLAUDE.md |
| Cursor | .cursorrules |
| GitHub Copilot (VS Code, JetBrains) | .github/copilot-instructions.md |
Plus MCP for tools that speak it — Cursor, Claude Code, VS Code Copilot Chat. The MCP server is wired by infernoflow setup / init into each tool's config file. No per-tool setup.
When the MCP server is wired, your AI agent can call these directly in chat:
| Tool | What it does |
|---|---|
amp_write | Log an entry (type, msg, optional file / line / tags) |
amp_read | Read entries with optional filters |
amp_search | Keyword search across entries |
amp_handoff | Generate the handoff document for the next AI session |
amp_health | Session health score (A–F) |
infernoflow_status | Memory + project health at a glance |
infernoflow_check | Validate the capability contract (read-only) |
infernoflow_context | Generate AI-ready context for a task |
infernoflow_git_drift | Detect which capabilities recent commits affected |
The amp_* tools follow the AMP MCP spec §7.3 — vendor-neutral. Any AMP-Full client only needs to know those five names.
infernoflow was developed by building a multi-tenant kanban (infernotest_01) and capturing what it surfaced. A sample of real entries from that dogfood:
vite.config.ts): "Vite proxy with changeOrigin: true rewrites the Host header — server-side URL construction produces URLs pointing at the BACKEND port. Build user-facing URLs client-side via window.location.origin."server/prisma/schema.prisma): "Prisma 6 query_engine.dll.node is locked while tsx watch is running; prisma migrate dev fails with EPERM on rename. Stop the dev server before migrating."server/src/routes/members.ts): "Invite accept must NOT burn the token when the caller is already a member of the workspace — return early with the existing membership before marking acceptedAt."server/src/routes/columns.ts): "Position assignment for ordered children: next position = max(existing) + 1024. The 1024 step leaves room for ~10 inserts between two siblings without renumbering."server/src/access.ts): "Cross-entity auth helpers do where: { memberships: { where: { userId } } } via Prisma nested-select — one DB hop per assertion. Return 404 not 403 when not a member to avoid leaking existence."server/src/auth.ts): "Opaque session tokens in a Session table (not JWTs) — chosen so we can revoke per-session (deleteMany on Session). bcryptjs over native bcrypt to avoid platform-specific binaries."These are the things you'd otherwise forget by next Tuesday and re-derive at 11pm on a Friday. They live in .ai-memory/branches/*.jsonl forever.
The companion extension renders your memory as a live sidebar — ranked-by-relevance gotchas/decisions for whatever file you're editing, status bar health score, inline CodeLens annotations at gotcha locations.
ext install infernoflow.infernoflow
Or in the Marketplace: infernoflow.infernoflow. Activates on any project with .ai-memory/ (or legacy inferno/).
The extension is window only in v0.7.9+ — the CLI is the single canonical writer of rule files. No race between extension and CLI; the extension watches .ai-memory/**/*.jsonl and renders.
amp_write entries still look wrong. Your IDE's MCP server is loaded into memory at session start and doesn't reload from disk. Quit and reopen Cursor / Claude Code / VS Code. infernoflow doctor will flag this with a "MCP runtime v… but CLI v…" warning.infernoflow not found. Use npx infernoflow until the global install resolves on your PATH.Set-ExecutionPolicy -Scope Process -ExecutionPolicy Bypass.infernoflow doctor — full diagnostic if anything looks wrong. Includes the MCP runtime stamp check + AI provider detection + git hooks status.Code changes daily. What the system actually does under all those edits — the invariants, the constraints, the things that bit you last week — code can't tell you. infernoflow keeps that current and feeds it to the agent so the agent stops re-deriving from scratch.
That's the whole product. No vendor lock-in (it's JSONL on disk). No SaaS. One CLI, one MCP server, three rule files your IDE was reading anyway.
MIT
FAQs
Persistent memory for AI coding sessions — captures what agents can't infer from code alone. Works with Copilot, Cursor, Claude, and Windsurf.
We found that infernoflow demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Research
/Security News
The North Korean malware loader hides in a Packagist-listed package and its GitHub branch to fetch and execute remote code in a likely Contagious Interview-style lure.
Research
/Security News
A malicious NuGet package impersonating Sicoob exfiltrated client IDs, PFX passwords, and banking certificates through Sentry telemetry.
Security News
Feross Aboukhadijeh joins TBPN to discuss Socket's $60M Series C, 500%+ ARR growth, AI's impact on open source, and the rise in supply chain attacks.