Persistent memory for AI coding sessions. Captures decisions, gotchas, and context — the stuff that lives in your head, not your codebase.
Works with your AI coding tools
infernoflow runs silently in the background, capturing what your AI can't infer from code alone. Local-first — no network calls unless you opt in to telemetry.
Log decisions, gotchas, and in-progress work. When you start a new session, your AI loads the full context automatically.
Say bookmark this before you stop — or before a risky change — and resume exactly where you left off next session. On Claude Code it distills the transcript automatically, no model call.
On Claude Code, an installed infernoflow-memory skill makes your AI log gotchas and drop bookmarks on its own, and a memory-keeper agent can sweep a session and capture what matters. Cursor & Copilot get the same guidance via their rule files + MCP.
On Claude Code, a session-start hook loads current memory into every new session, and a session-end hook leaves a resume point. infernoflow resume answers “where were we?” in one call.
Run infernoflow switch to generate a full handoff — open threads, git activity, session duration, and next steps.
Memory stored as plain JSONL in .ai-memory/ (per branch, tracked in git) using the AI Memory Protocol — vendor-neutral, MIT-licensed. Any AMP-compatible tool can read it. No lock-in.
infernoflow tracks what your codebase can do — not just what it contains. MCP tools keep your AI aligned with the contract.
Ships with MCP tools your AI calls automatically — amp_read, amp_write, amp_resume, amp_bookmark and more — registered per project. Only the read-only tools are pre-approved.
Three commands and your AI has persistent memory for every project.
Run npm install -g infernoflow once. Works on Node 18+.
Run infernoflow setup in your project root. Creates the .ai-memory/ folder, registers the MCP server for this project, and wires your AI tools: rule files for Cursor & Copilot, session hooks and a memory skill for Claude Code.
Run infernoflow log "decided to use Zustand over Redux" to capture decisions — or let your AI log gotchas and dead ends itself. Your AI reads it all on the next session start.
Run infernoflow switch to generate a rich handoff. Open threads, git activity, session duration, IDE — everything the next session needs to hit the ground running.