01
Learn
Your model fixes a failing test. Reeflex keeps the fix, then proves it in a sandbox: it has to transfer to renamed and restructured variants with zero regressions before it goes live.
Gated before it ships
Fix it once.Never twice.
Verified fixes forcoding agents, minus the model call
Reeflex learns the fixes your agent's model makes and replays them on the next similar failure. Private beta.
Every coding agent pays its model to rediscover fixes it has already made. Reeflex remembers them, checks them, and hands them back for free. Here's how:
Not a cache. A memory for fixes.
A cache replays the same request. Reeflex carries what your model learned to failures it has never seen before.
02
Transfer
The next time a similar failure shows up, in another file, with other names, Reeflex writes the fix itself. Within a session too: when an upgrade breaks 30 call sites, the model fixes one and Reeflex does the rest.
15 of 17 bugs, no big-model call
03
Verify
Your tests are the judge. Misses fall through to your model as usual, free. A patch that fails is never billed, and every promotion can be rolled back.
0 bad fixes across 26 bugs
Your agent keeps paying for the same fix.
The problem
Coding agents hit the same failures over and over: the off-by-one in a pager, the empty queue that returns undefined, the null nobody checked. Each time, the agent sends the whole context back to the model and pays for it to think the fix through from scratch.
The model gets it right. It got it right last week too. Nothing about that first fix is kept, so the bill grows with every task and the agent never gets faster at problems it has already solved.
Reeflex keeps the fix, proves it, and makes the next one free.
Models are brilliant. They just don't remember.
What Reeflex changes
A frontier model is the right tool for a failure nobody has seen before. It is an expensive way to fix the fiftieth copy of one you already have.
Reeflex sits in front of the model. When a turn matches a fix it has learned and verified, Reeflex writes the patch itself in milliseconds. Everything else goes to your model exactly as it does today.
Your model, for
Failures it has never seen, and anything Reeflex isn't sure about.
No longer paying for
The same fix again, at full price, every time it comes back.
If Reeflex is ever down or unsure, it steps aside. Your agent always runs.
Watch. Learn.
Transfer. Verify.
We replayed 57 real upgrade commits from public repos. Reeflex wrote 61% of the edits, the agent made 78% fewer model calls, and all 57 still passed.
01Watch
Reeflex sees your agent's normal traffic: the failing code, and the fix that turned the test green.
02Learn
The fix becomes a capability only after it passes every gate in a sandbox.
03Transfer
A similar failure arrives. Reeflex writes the patch itself, with no model call.
04Verify
Your tests decide. You pay only for fixes that pass.
Same 26 bugs. 77% less spend.
Real-agent replay
Measured, not projected. We ran Claude Code on 26 real bugs, with and without Reeflex: $5.99 became $1.39, and 4.5 million tokens became 1.8 million. All 26 tasks were solved, and not one Reeflex fix failed and fell back to the model.
We ran it on Codex too: 55% fewer tokens on the same real bugs, and Reeflex solved one the agent alone didn't.
Upgrades are where it adds up. Across 57 public commits that moved off datetime.utcnow, new Buffer and io/ioutil, model calls fell from 1,329 to 287. When a Reeflex edit failed the check, Reeflex put the file back and the model took over. Underneath those numbers are four promises:
How it stays safe01Fail-open
Any Reeflex error is a miss. Your agent's normal path always runs.
02Shadow mode
Start by measuring what would have hit on live traffic. Nothing is served.
03Yours only
Fixes learned in your project are only ever served to your project.
04Reversible
Every promotion is logged. Retire one, or stop everything, in a click.
One line to plug in. Nothing to break.
Integration
Reeflex speaks the OpenAI and Anthropic APIs. Change your client's base_url to Reeflex, keep your own provider key, and every call goes through exactly as before. Reeflex only steps in on the turns it can prove it knows.
Rolling it out is three steps, and you can stop at any of them:
- 01Point. Set base_url to Reeflex. Chat Completions, Responses or Messages, streaming included.
- 02Shadow. Reeflex records what it would have answered on live traffic, and serves nothing.
- 03Active. Reeflex serves verified fixes. Misses still go to your model, free.
You pay per verified fix. Misses and shadow decisions cost nothing.
Works with the agent you already run.
Reeflex is a drop-in gateway, not a new framework. If your agent talks to an OpenAI or Anthropic endpoint, it already works with Reeflex.
Claude Code
Set ANTHROPIC_BASE_URL to Reeflex and keep working. Reeflex fixes through Claude Code's own edit and Bash tools.
Codex CLI
Point Codex at Reeflex and run it as usual. Reeflex answers the turns it already knows over the Responses API.
OpenAI and Anthropic SDKs
Any client that takes a base_url. Same calls, same models, your own provider key.
OpenCode
Set the Anthropic provider's baseURL to Reeflex. Its edit, patch and multiedit tools all count.