Welcome to the future of programming.

An agentic harness and IDE, with the engineer back in charge.

Traditional programming doesn't scale.

You are falling behind.

Vibe coding gave you speed.

And took your control. Bugs in production. Nobody knows where to look. What shipped was never what the business asked for.

The new order: spec-driven development.

Supersonic speed, with control. fAI provides a framework to help you plan, decide, document and test efficiently.

A harness engineered for any model.

Claude, Codex, Kimi. Swap the engine, keep the process. Controlled delegation with a chain of command, not a swarm let loose.

Context that doesn't rot & token efficiency.

Just-in-time injection, automatic pruning. Smart navigation and automatic anti-staleness for your documentation and codebase.

Built in +15 agents · +80 tools · +90 skills

Licences from $49 / month · 30-day refund

From chaos to command. Agentic harness & IDE. All in one app.

You set the degree of autonomy for each occasion. Grab the wheel or let fAI manage the loop.

See pricing

The method

Control and flexibility,
at the same time

Each one answers a specific problem in working with agents: context that degrades, an interface that falls short, autonomy without a framework, documentation that ages, and work chained to the desk.

01
Frontier context engineering

Just-in-time context, with different lifespans and hierarchies.
- Every session gets a structured context sidebar for the learnings and decisions made inside it.
- Automatic promotion to longer-lived context (documentation, memories, skills) when something looks useful for future sessions.
- Automatic, configurable pruning of tools within the session, stretching it to 80 turns without compacting while keeping coherence.
- Context injected automatically when a new milestone is marked in process.
02

A clear, complete UI for the developer

This is not about an agent launching a swarm of 100 agents with no framework. It is about an engineer parallelizing work efficiently across 10 orchestrators, each delegating in an orderly way to N sub-agents, with a clean interface that lets them make decisions on several levels, fast and reliably.
03

Process engineering, not a loose swarm

The balance between the model's creative power and the control of an enforced framework. An orchestrator, a supervisor and specialist agents, each with its own tools and its own way of reasoning, designed for consistent results, all running inside a deterministic orchestration. An automatic loop driven by an approved technical spec every agent can read, with a supervisor agent and tests run automatically by the harness that decide whether a task is closed or not.
04

A second brain that evolves

Its own vault with a doc-to-doc and doc-to-code graph, which alerts the agent to code and documentation that went stale after a change. You write, comment and edit on the same living document: you, your team and the agent.
05

All of fAI from your phone

Scan a QR code and carry on from your phone while your desktop keeps working at home. No VPN, no network setup. It is not a read-only view: it is the full workbench.

The index

Everything that comes inside

Ten fronts, in the same window you work in. What follows is a selection. There is more inside.

01 Mobility and accounts

Access from your phone — the full fAI workbench, on your phone.
Voice — reads answers aloud, and you can dictate instead of typing.
Multi-account with automatic rotation between accounts.
Limits indicator — real usage against the pace you are burning through it.

02 Processes and performance

Long-lived processes — they outlive the agent that started them and are managed from fAI, with their logs visible.
Log pattern alerts — fAI tells you when what you were waiting for shows up, without polling.
Native Rust backend — indexing, memory and persistence run once for all agents, not one copy each.
Automatic memory control — idle agents hibernate and release their memory when the machine nears its limit, then resume where they were.

03 Context, threads and memory

Sessions sidebar — several threads and sessions in parallel, each in its project, with its history, context and labels. All in one place.
Context sidebar — the thread's context, live and editable: goal, tasks, decisions and learnings.
Automatic pruning — drops what no longer helps and stretches the session to 80 turns without compacting.
Rewind — go back to an exact point in the session. fAI snapshots the code before every agent edit and restores the files to that moment.
A built-in memory system, with three loading modes depending on how long each thing lives.
Vectorized history and documentation — complement graph search and rescue content buried in old threads.

04 Orchestration and subagents

Pre-built specialists for the key trades: debugging, implementing, researching, testing and querying project memory.
Parallel work — disjoint tasks, each in its own copy of the repository, integrated automatically when done.
Council of advisors — five distinct perspectives on the same decision.
Automations — triggered by event or schedule, with subagents packaged for the task.

05 Knowledge

A built-in skills system that loads a skill automatically when it applies: around fifty included.
Knowledge inbox — what the project learns is proposed there, and you review it before it gets written down.
Documentation promotion at different levels: project, user or global.
An editor for all your Claude config from the app — subagents, skills, commands, MCP, CLAUDE.md, env, permissions, plugins.

06 Artifacts and work tools

Canvas — a markdown document co-edited by you and the agent.
Design canvas — your frontend running inside fAI: point at an element on screen and the agent edits the code that produces it.
Change control — review what the agent touched piece by piece, and accept or reject each one separately.
Kanban with cards that can also trigger work.
Diagrams generated by the agent inside the conversation.

07 Planning and systems

System creation and visualization with a structured form + DAG (System graph).
Full planning with form + DAG (Plan board / graph).
Architecture decision catalog — cards with theory, pros and cons, and animated diagrams.
Export the technical brief to Markdown / Word / Notion, with real tables.

08 Chat and interaction

Answers with decision cards — each path with its advantages, costs and effort, so you choose with the trade-offs in view.
Dangerous-action guard — intercepts what cannot be undone, asks for confirmation and runs it only once you accept.
Correction on the fly — redirect the agent while it works, or leave the message queued for when it finishes.
Quote fragments — select part of the conversation and flag it for the agent to address.
Conversation minimap — jump to any point in the history.

09 Interface and navigation

Integrated file explorer and git — with structural code search by symbol, the same engine the agent uses.
Integrated terminal, with several tabs.
Quick open and command menu — reach any file or any app action from the keyboard.
Per-thread settings — model, effort level and planning mode, different for each conversation.
Linear integration.

10 Data and persistence

Local persistence — all your state lives in a database on your own machine: threads, sessions, context, memories, board and decisions. It never travels to the cloud.

Before you install

Three things need to be on your machine

fAI does not bundle what you already have. It is lightweight on purpose, and tells you at startup if anything is missing.

Windows 10/11 or macOS

64-bit Windows, or macOS on Apple Silicon.

Claude Code

The CLI installed and signed in. fAI wraps it, it does not replace it.

Node.js

Version 18 or later. Needed by the tools the agent plugs in.