Five Qwen3.8-27B agents beat Fable 5 on LiveCodeBench Hard, as a Memdoor workflow.
GVS5H's harness rebuilt with the authors' exact prompts and loop,
on the first ten problems of their list, five passes each: a mean of 9.8 of 10, against 9.6 for the
authors' own harness and 8.0 for Fable 5 in a single call. All 50 runs are published: every request,
reply, program and grader output. The workflow is YAML and names no model: point it at yours, or at a
local one with memdoor connect local (Ollama, llama.cpp, vLLM, LM Studio; how).
examples/workflows/team-solve · the HN thread
On this machine:
curl -fsSL https://memdoor.ai/install.sh | bash
export OPENROUTER_API_KEY=sk-or-... # or ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, …
cd your-project && memdoor # the first run sets the machine up
Or on a box you can ssh to, with nothing installed there:
curl -fsSL https://memdoor.ai/install.sh | bash # here, once
memdoor --gateway ssh://you@box # installs memdoor on the box, starts its gateway, opens the window on it
memdoor --gateway ssh://you@box connect openrouter # your key, kept on the box; every turn there runs on it
memdoor --gateway ssh://you@box run "fix the failing test" # the same box from a script or CI
The runs live on the box: close the laptop, the work goes on, and your phone or the next terminal opens the same conversation there.
No account. Telemetry off unless you turn it on. macOS, Linux, Windows, one static binary, Apache 2.0.
In GitHub Actions
One step in a workflow, on your key, with nothing to add to your repo
(Marketplace):
on: pull_request
jobs:
review:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v5
- uses: guregodevo/memdoor-action@v1
with:
workflow: pr-review
openrouter-api-key: ${{ secrets.OPEN_ROUTER_API_KEY }}
Nine are built in:
pr-triage sorts a pull request for a maintainer: ready for review, needs work, or likely slop, from checks run in the job (the tests, claimed tests that are not there, functions nothing calls).pr-review reviews the pull request, verifies each finding by running the code, and fails the job only on a blocker.security-review does the same for security only: secrets, injection, path traversal, missing authorization.write-tests writes tests for what the change touches, and fails the job when one shows the change is wrong.pr-description writes the description: title, What, Why, How to test, Risk.explain-failure reads a failed run's logs and the diff: the cause, file:line, and the fix.issue-triage suggests labels, the kind, a summary and duplicates for a new issue.fix-tests finds the test command, fixes the code until the tests pass, and leaves the fix in the working tree.release-notes turns the commits since the last tag into Added / Changed / Fixed.
No key at all: local-model: Qwen3-8B-4bit instead of the key runs any of them
on a model inside the runner (janis, one
file, the model cached), in one model call. It is a free first pass, not a
reviewer: on a free runner it caught an obvious and a subtle bug in about a
minute each, and can file a correct change's own fix as a blocker.
Or workflow: <name> for your own, .memdoor/workflows/<name>/ in the repo. The
job ends with the run's outcome (0 done, 2 failed, 1 stopped, 3 stopped at an
approval gate) and its summary shows the review, the fix or the notes. Trigger on
pull_request, never pull_request_target: GitHub gives a fork's pull request no
secrets, so a stranger's code never runs beside your key.
On a ChatGPT Plus or Pro plan? Skip the key: memdoor connect chatgpt (or
/connect chatgpt in the window) opens Sign in with ChatGPT, and your plan's
allowance answers Memdoor's turns. A Claude subscription cannot be used this
way (Anthropic's terms); Claude needs an API key.
No key at all? The first memdoor signs you in with OpenRouter in the browser,
a free account included (or with ChatGPT), from a gateway on this machine, over
ssh or on a box. An account with no credit runs on a free model picked from
OpenRouter's list, and the connect says so: 50 requests a day, and the free hosts
train on what they are sent, so a trial, not private code.
Or no key at all: memdoor connect local finds a model running on this machine
(janis, Ollama, llama.cpp, LM Studio),
and nothing leaves it. janis is ours: one Go server that runs the open models on
a Mac's GPU through MLX or on an NVIDIA box through CUDA, serves chat and the
decision model with no key, and runs Qwen3-Coder-30B-A3B on a $0.71-an-hour
Google L4; memdoor --gateway ssh://you@box connect local makes that box yours.
And that box answers your laptop: memdoor peers add box http://box:18789 --dir ~/app names it, and @box:coder … in any conversation here runs the turn there,
on the box's own model, and brings the answer back. Recorded on memdoor.ai: a
laptop with no GPU asks, Qwen3.8-27B on an L4 through janis reads, patches and
runs a Go program on the box, the window on the laptop shows it land.
Every run is a conversation on a gateway on your machine, whatever door it
came in by: a terminal window, memdoor run from a script or CI, a cron
schedule, your editor, or the page on your phone. The run a CI job started is
the one you open with memdoor resume <id>, watch live, and answer. One
gateway, every door (the walk through them, with exit
codes).
memdoor run "fix the failing test" # one turn, no window; the exit code is the receipt of its checks
memdoor workflow run release --wait # a workflow from a shell or CI, exit code the run's outcome
memdoor resume --last # the same run, in a window
memdoor reads your files, writes patches, runs the build and fixes what
it broke, in the directory you launch it from. Every model call goes from your
machine to the provider you connected, on your account, at their list price.
Two more things set it apart.
A decision model in front of the chat model. Jev (TypeSafe) answers a
typed question such as is this hit relevant to the task? with a calibrated
probability in under half a second, and the harness acts on it: search, file
and log reads return only what counts; only the tools the turn needs are sent;
a turn that stops making progress ends instead of looping to a cap. Measured
over 44 paired runs, decisions off and on: 22 of 22 passed each way; a question
about the codebase went from 54,661 to 14,367 input tokens (−74%); an edit with
a test and ten one-file fixes read the same files either way (within ±11%)
(docs/features/DECIDE.md).
Workflows. Say the steps and the coder writes them as task files, one YAML
per step, run as a graph by mario.
Independent steps run in parallel; a step is done when its check passes, never
because the model says so; a gate waits for your approval with the diff in
front of you; a failed run resumes at the failed step. The files live in your repo, so a workflow is kept, shared and
run again, by hand or on your local cron
(docs/features/WORKFLOWS.md). Why workflows
rather than skills: the write-up.
Worked examples, each run before it was added: examples/workflows/.
One of them, the step that grades a trading rule (examples/workflows/nightly-rule/steps/grade.yaml), as it runs nightly; no code in it:
type: agent
prompt: |
Today's partition is {{.partition}}.
Read rules/{{.partition}}.md and grade today's trading rule: assess it
briefly (clarity, risk profile, plausibility) and write a JSON object
{"score": <0-100 integer>, "verdict": "<one sentence>"} into
grades/{{.partition}}.json (create grades/ if needed).
requires:
- table_pattern: build
target:
# run by the engine, not by the agent: the grade must exist and be newer than step 1's start stamp
command: test "grades/{{.partition}}.json" -nt "rules/{{.partition}}.start" && test -s "grades/{{.partition}}.json"
timeout: 10m
A recorded session of the review loop (findings verified by running, a gate,
the fixes applied) plays at memdoor.ai/workflows.
Install
curl -fsSL https://memdoor.ai/install.sh | bash
export OPENROUTER_API_KEY=sk-or-... # or ANTHROPIC_API_KEY, OPENAI_API_KEY, GEMINI_API_KEY, … or /connect inside
cd your-project && memdoor # the first run sets the machine up
macOS (Apple Silicon and Intel), Linux (amd64, arm64; static), Windows
(irm https://memdoor.ai/install.ps1 | iex; the agent runs commands in
Git Bash when it is installed, otherwise PowerShell). One binary, installed to
~/.local/bin, no sudo, no Docker, no Python. The installer
(scripts/install.sh) verifies the SHA-256 of what it
downloads.
From a clone, with Go 1.25:
git clone https://github.com/guregodevo/memdoor && cd memdoor
make build # ./memdoor, stamped with the tag it was built from
cp memdoor ~/.local/bin/ # or anywhere on your PATH
In the window
| Command | What it does |
|---|
/model | Which model answers: the agent's ladder (cheapest rung first), or pin any model of your providers, with list prices |
/connect | Add a provider: OpenRouter, Anthropic, OpenAI, Gemini, DeepSeek, Baseten, Groq, xAI, any OpenAI-compatible endpoint, your company's AI gateway; /connect local a runtime on this machine or the box (Ollama, llama.cpp, janis), no key; /connect chatgpt signs in with a ChatGPT Plus/Pro plan, no key |
/workflow | The project's workflows and runs; enter opens a run's graph, a approves a gate, /workflow:<name> runs one |
/usage | This month on your key, per model, and what the decision model kept out of the bill |
/remote | This conversation on your phone: a link and a QR code, the terminal itself, end-to-end encrypted |
/remote view | A watch-only link: whoever opens it sees the terminal live and cannot type, no account, up to 200 at once; ?live=1 is the terminal alone, for a stream or OBS |
/share | A read-only link to the conversation, encrypted, secrets removed; /unshare deletes it |
/files or ctrl+f | The project's files, the ones this conversation read or changed first, each with what happened to it; enter previews, ctrl+d diffs, ctrl+e opens in $EDITOR, tab puts @path in the prompt |
/graph or ctrl+g | This turn as a graph: the reads, the edits that depended on them, the runs that depended on the edits, the failing check in red |
/mcp | Connect MCP servers from a URL, a command or a .mcp.json snippet |
/help | The rest |
Esc interrupts a turn, ctrl+f opens the files panel, ctrl+g the turn as a graph, ctrl+o expands tool
frames, Shift+Tab sets the reasoning effort, @path mentions a file. MEMDOOR_APPROVE=changes asks
before every command and write, for machines whose policy requires it.
In your editor
memdoor acp serves the same coder over the Agent Client Protocol:
Zed and JetBrains host it natively; VS Code through an ACP client extension
(for example "ACP Client"), with one setting:
"acp.agents": { "Memdoor": { "command": "memdoor", "args": ["acp"] } }
Each editor session is a conversation on your local gateway; a prompt is a
turn in the editor's workspace folder, and the turn's text, tool calls and
approvals show as the editor's own. The steps: memdoor.ai/docs/editor.
From a script
memdoor run "make the retry wait configurable and add a test" # a headless turn, no window
git diff | memdoor run "review this diff; one line per finding" # stdin is the input
memdoor run --json "fix the failing test" | jq 'select(.event=="done")'
memdoor workflow run release --wait # progress on stderr, results on stdout
memdoor remote --last # the link /remote gives, on stdout; the gateway holds the run for your phone
memdoor remote --view --last # the watch-only link: an audience, or a pair of eyes
The exit code is the turn's receipt: 0 checked, 2 unverified or a failing check,
3 a question nobody answered (--yes answers it), 1 an error, 130 interrupted. The gateway
starts if none is running; every run is a conversation memdoor resume opens.
Headless, then attached.
A workflow in GitHub Actions is one step: In GitHub Actions above.
Where your data goes
Files, sessions and commands stay on your machine. What leaves is what the
model reads, sent to the provider you connected; on OpenRouter every request
carries data_collection: deny and hosts that train on paid inputs are kept out
of the ladders. The one exception is a :free model, one you pin or the one an
OpenRouter account with no credit answers on: its hosts train on what they are
sent, the request says so (allow), and Memdoor says so at the pin or when the key is connected; use one for a trial, not
on private code. On a vendor key or a company gateway nothing else is contacted:
docs/SECURITY.md lists every host the binary can name and
a test fails the build on a new one.
Documentation
Repository layout
cmd/cli/ the CLI (cobra): tui, model, workflow, cron, account, …
gateway/ the local gateway: agent runtime, tools, providers, workflows
pkg/ the domain: decision, metering, plan, workflow (the mario task factory), …
web/ the memdoor.ai site and the served docs (React, Vite)
scripts/ install.sh, install.ps1, deploy.sh
docs/ features, reference, ADRs
examples/ workflows built by the coder and run, to copy into a project
License
Apache License 2.0 — LICENSE, NOTICE. The public
repository is github.com/guregodevo/memdoor; issues there.
