The command center for serious AI work.
Open Free Max brings your projects, interactive AI sessions and repeatable workflows into one local desktop workspace. For vibe coders, creators, freelancers, researchers and builders who have outgrown one project and one prompt at a time.Use the official agents and subscriptions you already have - with Mission Control, local memory and direct Obsidian access.Kimi CLIs interactively — all from one Mission Control
Free · Windows v4.5.0 (.msi) · macOS v4.5.0 (.dmg)
Open Free Max uses up to 50× less memory than Antigravity, Cursor or Visual Studio.
New in 4.5.0
Browse the official agent library: specialized, reusable workflows with a clear purpose, their required capabilities, and installation guidance. Choose an agent, review its compatibility, then install it in the project where you need it.
Installation is verified and project-scoped, so your other projects stay untouched. See the4.5.0 changelog for every change.
Not another chatbot window. Not a fragile pile of terminal tabs.A calm local workspace for the AI work you already do - with the control, memory and automation to carry it further.A thin native shell over the official agent CLIs —
Your own local knowledge base.
Like Obsidian, OFM turns scattered context into durable, connected knowledge. Every project gets its own local base of notes, facts and links, with semantic search and human review. Your agents can use it across sessions and CLIs, and it never leaves your machine.
Signed entitlement so the client can't be pointed at a fake backend. Public key burned in at build time…
Access tokens stay HS256; only the entitlement is RS256-verified client-side…
The Obsidian-like knowledge layer built into your IDE.
OFM builds a private knowledge base for each project — notes, linked facts, decisions and provenance — and lets you search it instantly or make it available to the next agent.
- Search engine over your history — semantic search across everything your agents did in a project, with one-click resume of the original Claude session
- Distilled facts you approve — OFM turns sessions into readable facts; you curate what sticks, then it auto-injects them into the next agent (MCP for Claude, a tagged block for the others)
- Shared across CLIs — history read from Claude, Codex, Antigravity and Vibe, so memory follows you between tools
- Notes and links, not a black box — inspect the knowledge, trace its source and decide what your agents should retain
- 100% local, private by design — the memory is a single SQLite file on your disk. OFM never collects, uploads or even sees your exchanges with the LLM — no account, no cloud, no API
Connect your live vault directly through MCP.
OFM configures Claude Code or Codex to use the MCP server built into the Obsidian Local REST API plugin. Your agent can search, read, create and update notes in the vault currently open in Obsidian. The connection stays on localhost, uses bearer authentication, and keeps the external vault separate from OFM's native project knowledge base.
Your conversations never leave your computer.
Memory is computed and stored entirely on-device. Open Free Max does not collect, transmit or retain your prompts or the agents' replies — there is no telemetry of your LLM exchanges and no server to send them to. Delete memory.db and the memory is gone, instantly.
Start with an agent that already knows the job.
Browse specialized, reusable workflows instead of rebuilding the same operating instructions for every project. OFM shows what an agent does, what it needs, and whether it fits before you install it.
Useful starting points, without the guesswork.
Pick a focused workflow for the work in front of you — from technical SEO and citation readiness to content systems and delivery quality. You review the fit; OFM handles the safe setup.
- A clear purpose for every agent — its outcome, capabilities and operating guidance are visible before you choose it
- Compatibility checked up front — spot missing CLI support or required capabilities before an installation begins
- Verified, project-scoped installation — add the selected agent to one project without changing the rest of your workspace
Technical audit · indexing watch · citation readiness
Evidence-led findings, local fixes, and a validation plan.
Describe the task. It picks the agent.
Stop picking a model before every task. Describe the work. The right agent starts, at the right level of thinking — and you see the choice before anything runs.
One decision less, dozens of times a day.
You own several subscriptions and several models. Nobody re-picks the best one for every little task, so everyone settles on one and overpays or underthinks all day. The router picks per task. You keep the last word.
- A typo no longer costs you a heavyweight model, and a hard question no longer gets a light one
- No more Friday wall — when one subscription runs out, unattended work carries on with another one you already own
- You stay in charge — the choice is shown before launch, and one click sends the task to another agent
- Several opinions on a question that matters, without paying more — your plans are already paid
- It picks up your habits on each project — and one button makes it forget them
Model: opus · one click to run it elsewhere
Every agent you're running, on one screen.
Spin up as many CLI sessions as you want — across projects and providers — and watch them all from a single control room. Each pane is its own live PTY: glance at what every agent is doing, jump in to steer one, and let the rest keep working.
- Unlimited parallel sessions — Claude, Codex, Antigravity, Cursor, OpenCode, Mistral Vibe, Kimi, side by side
- One glance shows what's running, idle or done across every project
- Click any pane to take over its prompt — the whole layout is restored next launch
Work that keeps happening without you.
Give an agent a checklist and a schedule. OFM wakes it on its own, lets it act if there is something to act on, and acknowledges in silence when there isn't. As many recurring jobs as you want, across as many projects as you want — running alongside the sessions you drive by hand.
A checklist, a schedule, and an agent that reports only when it matters.
Every turn ends one of two ways: a report you can open, or an acknowledgement meaning "nothing to do". The status line updates either way — but only real work becomes something to read.
- Interval or fixed times — every N minutes, or at set hours of the day, with an optional active-hours window in your machine's own time zone
- Silence is the default outcome — an acknowledged turn writes no report, so notifications stay meaningful instead of becoming noise you filter out
- Ambiguity stops the turn — a genuinely undecidable question is never guessed at. It reaches your approval queue, tagged as coming from a recurring job, with a system notification
- Per-job CLI settings — approval flags, memory and vault access pinned to that job alone, so your careful interactive setup stays careful
- A session you open by hand always starts — immediately, even when every orchestrator slot is busy running turns
Nothing to do — no report written.
Two advisories in transitive deps, both patched. Lockfile updated, test suite green.
Major version bump would break two call sites — sent to the approval queue instead of deciding alone.
What it is not — stated plainly.
Recurring work does not make an agent smarter, and it is not a full cron. If your machine goes to sleep, no turn fires — OFM will not wake it, and a closed laptop lid still forces sleep. Missed slots are never replayed: a machine switched on after three days runs the next slot, not the 144 it skipped. That is deliberate, and we would rather you knew before you set one up.
Your agent. Your plan. No surprise API meter.
Providers split usage in two: interactive — a human typing and reading — runs on your normal subscription. Programmatic— SDKs, headless runs, third-party app protocols — drains a capped credit pool and then bills at API rates.
What OFM does
- Drives the official CLI in interactive mode (human in the loop)
- Renders its real stdin/stdout through an embedded PTY
- Leaves auth entirely to the official binary — keys never leave it
What OFM refuses
- The Agent SDK and headless
-pruns - ACP / third-party app protocols (metered as programmatic)
- Ever handling your tokens, keys or session auth
Everything an agent IDE should be — and nothing it shouldn't.
An AI Router across your subscriptions
Describe the task; OFM picks the CLI, the level of thinking and the model, and shows its choice before launching. One click sends it elsewhere. New in v4.0.
Up to 50× lighter
Built on Tauri — your OS WebView and a native Rust core. No bundled Chromium, no VS Code extension host. Up to 50× less RAM than Antigravity, Cursor or Visual Studio, even with several agents running.
On your subscription, not your API bill
OFM drives the official CLIs in interactive mode — a human in the loop. Your usage stays on the Max / Pro plan you already pay for, never the metered programmatic path.
Sessions that survive a restart
Close the app, reopen it, and every agent tab replays its full conversation. OFM never owns the transcript store — so a conversation is impossible to lose or corrupt.
First-class screenshot workflow
Paste or drop any number of images into a prompt. OFM writes them to disk and shows thumbnails — you never type a file path, and clipboard paste just works on native Windows.
Your local knowledge base
OFM has its own Obsidian-like knowledge base: notes, facts and links per project, searchable and available to your agents. It is stored 100% locally and never uploaded.
Direct Obsidian connection
Connect Claude Code or Codex directly to the vault open in Obsidian through the MCP server built into the Local REST API plugin. Search, read and write notes without copying the vault into OFM.
Mission Control
Pilot as many agent CLIs as you want from a single screen — every session live in its own PTY, across projects and providers. Steer one, let the rest run.
Recurring work that runs itself
Wake an agent on a schedule, hand it a checklist, and let it act only when there is something to act on. Nothing to report means a silent acknowledgement — not a notification you learn to ignore.
Usage, cost & savings, locally
OFM reads each CLI's own usage data and shows what you actually consumed per project and per agent — plus what the same work would have cost at API rates. Computed on your machine, never uploaded.
Native WSL2, no extension
Open a project living inside a WSL distro and the agent runs in Linux — node, git and the CLI from your distro, the right PATH guaranteed. Nothing to install on the IDE side.
Conversations you can't lose
The agent CLI already stores every transcript on disk. OFM keeps only a tiny map of which sessions were open — and rebuilds it from disk if it's ever missing. That's the bug other tools can't fix, fixed by design.
Screenshots, the way you work
Hit Ctrl+Vor drag images straight into the prompt. OFM saves each one, shows a thumbnail chip, and injects the real paths at send time. Drop ten — no path-typing, no silent clipboard failures.
The people building more with AI.Choose your workflow.
Open Free Max is for advanced AI users who run several projects and already know the value of interactive agents - vibe coders, creators, freelancers, researchers and product thinkers who need to produce, iterate and ship at scale without writing pipeline code.
The Vibe Coder
Ships products by steering outcomes, not terminal windows.
Describe the feature, the agent builds it. OFM keeps multiple build sessions live across your projects so you steer by outcome, not syntax. Hit your quota mid-sprint? Sessions park automatically and auto-resume when your plan refreshes - the build never stops.
The Creator
One idea becomes a complete content system.
Turn one editorial idea into a whole system - posts, threads, newsletter drafts - with several agent sessions running side by side on your own subscription. Project memory keeps the brand voice and the brief consistent across every piece.
The Freelancer
Every client gets a workspace that remembers.
Each client lives in its own persistent session - its own context, files, and conversation history. Switch projects without re-briefing. Close the laptop, return tomorrow: every session is exactly where you left it.
The Product Thinker
From research to a decision-ready deliverable.
Run UX synthesis, competitive analysis, decision writing and specification work as parallel agent sessions. Keep the reasoning in local memory or connect the Obsidian vault where your notes already live.
The Content Studio
500 deliverables. Not 500 manual prompts.
SEO pages, product descriptions, email sequences, localised copy - keep a session open per workstream and move between them from one screen. Still on your subscription, never the API meter.
The Course Builder
A complete programme, built in batches.
Keep the syllabus and the lesson template in project memory, then work through the programme with several sessions live at once. Hit your quota and sessions park, ready to pick up the next morning. You focus on the curriculum and the quality bar.
Works with the official CLIs.
One provider adapter, the agents you already use. No lock-in.
Claude Code
Anthropic — session resume, names, fork
Codex
OpenAI — interactive CLI
Antigravity
Google — interactive CLI
Cursor Agent
Cursor — interactive CLI
OpenCode
Open source agent CLI
Mistral Vibe
Mistral — interactive CLI
Kimi Code
Moonshot — interactive CLI
Thin where it counts.
Open Free MaxOne plan. Every agent you pay for.
Open Free Max is the IDE. Your agent usage keeps running on the Claude, Codex or Antigravity subscription you already have — OFM never adds an API meter. Start with a 7-day free trial, no credit card.
OFM Pro
Everything in Open Free Max.
- Mission Control — unlimited parallel agent sessions
- Obsidian-like local project memory, shared across CLIs
- Recurring work — scheduled agent jobs on your own subscription
- All providers: Claude · Codex · Antigravity · Cursor · OpenCode · Vibe · Kimi
- Bulletproof sessions, screenshot workflow, native WSL2
- Windows & macOS · up to 50× lighter than Electron IDEs
- Free updates throughout the 3.x line
Download OFM, sign in, and your 7-day trial starts automatically.
Questions
How do I install it?+
Download the installer for your OS and run it — a native .msi on Windows or a signed and Apple-notarized .dmg on Apple Silicon Macs. No package manager, build step or command line required.
How much does Open Free Max cost?+
OFM Pro is $19.90/month, with a 7-day free trial and no credit card. That's the price of the IDE itself — your agent usage keeps running on the Claude / Codex / Antigravity subscription you already have. See Pricing.
Do I need an API key?+
No. OFM never touches your tokens, keys or auth. It launches the official agent CLI, which handles its own subscription login. You sign in once where you already do.
How does billing actually stay on my subscription?+
OFM only ever drives the official CLI in interactive mode (a real human typing and reading). It deliberately avoids the SDK, headless `-p` and ACP paths that are billed against the metered programmatic credit pool.
Does OFM store or upload my conversations?+
No. The new persistent memory is computed and stored entirely on your machine, in a single local memory.db file. Open Free Max does not collect, transmit or retain your prompts or the agents' replies — there's no telemetry of your LLM exchanges and no server to send them to. You search your own history locally, and deleting the file wipes the memory instantly.
Can OFM work directly with my Obsidian vault?+
Yes. OFM connects Claude Code and Codex to the vault currently open in Obsidian through the MCP server built into the Obsidian Local REST API plugin. Your agent can search, read and write vault notes while OFM's own project knowledge base remains separate and local.
How does OFM decide which agent and which model runs a task?+
That's the AI Router, new in 4.0 and on by default. You describe the task; it returns the CLI, the level of thinking (quick task, regular work, hard thinking), where that decision came from, and the model — and you launch or send it elsewhere in one click. You can also name the level yourself in the brief, and the router takes it as given. See AI Router.
Does the router put my agent work on an API meter?+
No. The AI part of the decision is paid by OFM and never touches your subscription quota. The agent it picks runs as a normal interactive session on the subscription you already pay for, exactly like a session you start by hand.
Can an agent work on a schedule, without me watching?+
Yes — that's recurring work, new in 3.5. You write a checklist, pick an interval or fixed times of day, and OFM wakes an agent on your subscription to run it. If there's nothing to do, the job acknowledges silently; if it acts, you get a report. A genuinely ambiguous question stops the turn and lands in the approval queue instead of being guessed at. See Recurring work.
Does scheduled work bill against the API?+
No. A recurring turn is the same interactive CLI session OFM always drives, just started by a timer instead of by your click. Flags that would switch a CLI into programmatic mode (-p, --print, exec, --output-format) are rejected — a recurring job cannot open a metered path.
What happens when I hit my token limit?+
Nothing breaks. OFM detects the token-limit window, marks the session as rate-limited and keeps it in place. When your plan refreshes, you pick the conversation back up exactly where it stopped — the transcript lives in the CLI's own store, so nothing is lost in between.
Which platforms are supported?+
Windows (with native WSL2 project support) and macOS — the Tauri core covers both with a small, native footprint.
Why not just use Cursor or a VS Code fork?+
Those embed Chromium plus the VS Code extension host, which is heavy, and they re-implement session storage — the place where agent conversations get lost. OFM stays thin and leans on the CLI's own durable store.
The 4.5.0 release is live.
Both production installers are now signed and ready. Download Open Free Max for Windows or macOS and drive every agent you already pay for.
Free · Windows v4.5.0 (.msi) · macOS v4.5.0 (.dmg)