Cursor · Claude Code · Codex

Bridging intention to intelligence.

OptiFlume knows which intelligence to call, what context to carry, and where agents must stop. For Cursor, Claude Code and Codex.

On your machine. No proxy. No API keys.

Pro free for 30 days, no card. The Guardian stays free for good.

Real recording

A small edit, moved off a frontier model.

Claude Opus 5.5 was selected for a one-line request. OptiFlume switched the model under the composer to GPT-5.2 before the prompt went out.

Less guesswork. More intelligence.

From model selection to conversation memory and agent safeguards, OptiFlume handles the decisions behind your AI workflow so you can focus on the work itself.

Token optimizer

Efficient Router

The right model and effort for every prompt. Routine work goes to a lighter model, hard problems to the strongest one, so no tokens are burned on the wrong model. Every switch shows where you work, with a one-line reason.

Prices span about 100x from the lightest model to the strongest.

Context management & collaboration

Context Manager

Compactions and model switches stop wiping the slate. Goals, decisions, constraints and the approaches that already failed carry over, so the next model picks up where you left off.

Coming next: start a session in one tool and pick it up in another.

Safety, security and focus

Guardian

Keys, tokens, personal details, card numbers and bank details are kept out of what the model reads. Destructive commands are warned about or blocked before they run. Loops and sessions losing track are called out. Every intervention is logged on your machine.

Free on every plan.

Learns how you work

Adaptive Intelligence

It reads the code a task touches and how the session is going, not only your words. Every time you overrule a pick, it learns how you judge difficulty, so its choices drift toward yours.

On Pro. Your style stays on your machine.

Sound familiar?

Out of requests by the 20th?

Most of what you ask an AI coding tool to do is routine: rename this, write that test, commit. Sending those to a frontier model burns your allowance on work a lighter model does just as well.

It forgot everything after compacting?

Two hours in, the chat compacts and the model starts again from a summary: goals blurred, failed fixes tried again. Your context should outlast the conversation.

Watched an agent reach for rm -rf?

Agents read files and run commands on their own. One wrong command, or one key read into a chat, is all it takes. You shouldn't have to check every line to feel safe.

Going in circles at 1 a.m.?

The same fix, failing the same way, for the fourth time. It is hard to see from inside the loop. OptiFlume notices, and gives you the evening back.

Paying more to re-read the chat than to write the code?

In long agent sessions, most of the bill, often more than 80%, goes to carrying context: writing the conversation to the cache and reading it back, turn after turn. Model choice is the visible cost. Context is the one nobody is managing.

One install answers all five.

See how it works

Independent by design

Don't let the fox guard the henhouse.

Model providers, and the tools that resell their models, earn more when you use more tokens. That is their business. Advice on using fewer should come from someone who earns nothing from them. OptiFlume charges one flat price and takes no cut of your spend, so every call it makes is in your favor.

No black box.

  1. 01

    Always visible

    Every switch and every catch shows where you work, with a one-line reason. Nothing changes behind your back.

  2. 02

    You have the last word

    Choose not to route any prompt and keep your own model, pause OptiFlume for a project, or turn it off for a tool. How to steer it →

  3. 03

    Stays on your machine

    Every decision is made locally, from your code and your session. Your prompts and code are never sent anywhere.

  4. 04

    Fails open

    If anything in OptiFlume breaks, your prompt goes through unchanged on the model you chose. It can never lock you out.

  5. 05

    Easy to undo

    Logs you can read, settings you can change, and one command to uninstall.

›

Where it stands

In detail: vs Cursor Auto · vs proxy routers

Decides on your machine; never sends your code anywhere

OptiFlume
Yes
Cursor Auto
No, in Cursor's cloud
Proxy routers
No, prompts go through the router's servers first

Sees the live session: git state, edit loops, errors

OptiFlume
Yes
Cursor Auto
No
Proxy routers
No

Every switch visible, and yours to override

OptiFlume
Yes
Cursor Auto
No
Proxy routers
No

Learns your style from the prompts you overrule

OptiFlume
Yes (Pro)
Cursor Auto
No per-user tuning documented
Proxy routers
Varies

Manages context, not only model choice

OptiFlume
Yes
Cursor Auto
No
Proxy routers
Varies

How you pay

OptiFlume
One flat price, no cut of your spend
Cursor Auto
Each request at the routed model's list price
Proxy routers
Often a share of routed spend, or a markup per token

Available to individuals

OptiFlume
Yes
Cursor Auto
Yes; the newer Cursor Router modes are Teams and Enterprise only
Proxy routers
Varies

See what it did for you.

Your console shows your own usage, on Pro and during the trial: what ran, what OptiFlume switched or suggested, and, in Cursor, the spend it avoided against the model you had selected, across every Mac on your account. Counts only; never your prompts or code. Turn it off with one command.

  • Spend avoided in Cursor over the last 30 days, and per day
  • The models that actually ran, and what OptiFlume did with each prompt
  • Switches made, and the choices you kept that tune it to your style
  • Context compactions
  • Loops caught, secrets redacted, risky commands warned or blocked

Example: one developer in Cursor, 30 days

optiflume.ai/console

Spend avoided, example month

in Cursor, at published prices

$40.40

About $1.35 a day, 59 prompts a day.

1771

prompts routed

67%

switched to a better-value model

1180 switches

98

times you kept your own model

each one tunes it to your style (Pro)

23

long chats carried forward

working-memory handoffs

Models that ran

Claude Sonnet 5
36%
GPT-5.2
28%
Claude Opus 5.5
21%
Composer 2.5
9%
Claude Haiku 5.5
5%

How OptiFlume handled them

Switched for you
1180
Suggested a better-value model
64
Kept your model when it switched
41
Kept your model before sending
57
Left as it was
429

If you don’t believe it yet.

The questions engineers ask before installing, answered the way we would answer them in person.

All questions →
Which tools does it work with?+

Cursor, Claude Code and Codex, in their VS Code panels and in the terminal, with more tools on the way. It is on as soon as you install, and one license covers every tool on every Mac you sign in on.

Cheaper models write worse code.+

Not always. Good code comes from good context as much as from an expensive model. OptiFlume judges how complex each task is and sends it to the model it needs: hard work to the strongest model, routine work to a faster one. After your tool compacts a long conversation, the next model gets a short working memory of your goals and decisions (Pro), so it starts with the context it needs. When it is unsure, nothing changes.

Does my code go to your servers?+

No. Everything OptiFlume reads and decides happens on your machine, and OptiFlume never sends your prompts or code anywhere. (Cursor, Claude Code and Codex still send your prompt to the model you use, as they always have.) The routing engine works offline.

What if it breaks?+

It fails open. If any part of OptiFlume errors, your prompt goes through unchanged on the model you selected. It cannot lock you out of your editor. Uninstall is one command.

For teams, coming soon.

Team-wide visibility into what drives the invoice and per-project routing policy across Cursor, Claude Code and Codex, with the decision still made on each engineer's machine. Early access is open.

Talk to us

Start with your next prompt.

Create your account, install, and keep working. The first time OptiFlume steps in, you will see what it did and why. Pro is free for 30 days, and the Guardian stays free for good.

macOS 11+ · Apple silicon and Intel · Signed by OptiFlume, notarized by Apple · Never locks you out · Uninstall with one command