Troubleshooting
What woodls says when something goes wrong, in its own words, and what to do about it.
The guides
One command
It finds most problems.
woodls doctorIt checks this computer: the terminal, your account and its key, woodls’ server and the ledger,
the sandbox, the engine and your graphics chip, lending and your private groups. Each line says
what it found, and the ones that need you say what to type. What it finds stays on this computer.
Inside woodls code, /doctor does the same.
The tables below quote what woodls prints, so you can search this page for the words you see.
Once
Then never again.
| It says | What to do |
|---|---|
| woodls needs a Mac with Apple silicon (M1 or later), and this Mac has an Intel processor | woodls doesn’t run on Intel Macs |
| this terminal runs under Rosetta | Nothing: woodls is installed for Apple silicon all the same |
| hosting needs macOS 15 or later, and this Mac has macOS … | Chat and woodls code work; to lend, update macOS in System Settings, Software Update |
| woodls code’s commands, and hosting, run in a sandbox made with bubblewrap | On Linux, install bubblewrap from your system’s packages, such as sudo apt install bubblewrap |
| Windows needs a restart to finish turning WSL2 on: choose Restart, not Shut down | Restart; the installer opens again when you sign in, or run the same command again |
| WSL2 still isn’t on after the restart | Virtualization may be off in the firmware, or the PC’s organisation keeps WSL off. Fix that, then run the installer again |
| Windows wasn’t allowed to … so nothing changed | A work or school PC: ask whoever looks after it to run the command it shows as an administrator, once |
| WSL2 couldn’t start … | Restart Windows; if it says so again, turn on Intel VT-x or AMD SVM in the firmware settings |
| a network that checks HTTPS with a certificate of its own | A company network: its certificate must be added to Ubuntu, as its IT team can tell you |
On Windows, run the installer again with $env:WOODLS_DEBUG='1' first to see every step it takes.
It keeps a log in %TEMP%\woodls-install-*.log, with no password in it, to send us.
Opening it
And using it on more computers.
| It says | What to do |
|---|---|
woodls opens accounts by invitation during the alpha: woodls init --invite CODE | Use the code from your invite’s email, or ask for one. woodls code on your own key, and a friend’s group, need none |
| that invite code isn’t one woodls made | Check it against the email: woodl- and two groups of five letters and digits |
| that invite code has opened an account already, and each opens one | If it was you, on another computer: take the account back with your recovery file, or add this computer with woodls account add-computer there. If not, reply to your invite’s email |
no account on this computer yet: woodls init --invite CODE | Open one, or woodls account recover FILE with your recovery file |
| no recovery file yet: without one, a lost key means lost credits | Run woodls account recovery and keep the file off this computer: Your account |
| this computer’s key isn’t authorized for account … | On the computer that holds the account, run the woodls account authorize line it shows |
| this computer has used its daily limit for the account | It starts again at midnight UTC, or raise it with woodls account authorize KEY --daily N on the account’s computer |
| this computer’s authorization for the account has run out | Renew it with the woodls account authorize line it shows |
| this account may not use woodls’ public network for now | Write to hello@woodls.com if you think that’s a mistake |
The public network
Busy, quiet or out of reach.
| It says | What to do |
|---|---|
| no computer on the public network is serving a model right now | Try again later (woodls public shows the network now), or ask a private group of yours with --group |
| the computer answering is busy; trying again in 5 s | Nothing: woodls tries twice more by itself. Esc or Ctrl+C stops the wait |
| … isn’t served right now; asking … instead, at its own price | Nothing: your saved model is back once a computer serves it |
| not enough credits (spendable …) | woodls credits shows yours. Lending earns more, spendable once no longer pending; a private group is free |
| woodls’ server didn’t answer just now (it may be restarting): try again in a minute | Try again in a minute |
| woodls’ server (api.woodls.com) couldn’t be reached | Check this computer’s connection, then woodls doctor. A provider’s key, your own server and a model on this computer work without it |
| this account has sent as many requests as it may this minute: try again in … s | Wait that long |
| several requests in a row broke the acceptable-use policy, so this account waits … | Nothing was charged; see the acceptable-use policy |
| Your credits couldn’t be read | woodls’ server wasn’t reached: the line after it says why |
Its own words
And the fix.
| It says | What to do |
|---|---|
| woodls code needs a model, and found none here yet | Give it one: a provider’s key, Ollama or LM Studio running with a model, --server ADDRESS, or an account |
| there’s no Anthropic key here | woodls code connect anthropic, or set ANTHROPIC_API_KEY |
| OpenAI’s API answered 401 …; check its key | The provider refused the key: woodls code connect openai keeps a new one |
| … answered 404 …; woodls code models openai lists the models it has | The model’s name is wrong for that provider: pick one from the list |
| … the OpenAI account has no credit left | Add credit where you made the key, then try again |
| your server at … didn’t answer: nothing is listening there | Start the server, or check the address and port |
| –model foo/bar: foo isn’t one of the providers | Write the provider first, such as anthropic/claude-sonnet-5-5, or add --server or --this-computer |
| there’s no model on this computer now | woodls models get qwen3-8b downloads one, or use Ollama, LM Studio, a key or your server |
| woodls’ engine isn’t on this computer yet | On Linux or Windows, run woodls code --this-computer in a terminal and say yes to fetch it. On a Mac, install woodls again |
| woodls has no engine for this kind of computer yet | ARM64 Linux: run the model in Ollama or LM Studio, and woodls code finds it as your server |
| woodls’ engine needs a newer Linux than this one | It needs glibc 2.39 (Ubuntu 24.04, Debian 13, Fedora 40 or later). Ollama, LM Studio, your server, a provider and the woodls network still work |
| woodls’ engine runs only in its sandbox, and it can’t be made here | See the sandbox on Ubuntu 24.04, or use Ollama or LM Studio |
| … doesn’t fit on this computer beside what it has loaded now | Free some memory: woodls host --stop stops lending here. Or choose a smaller model |
| woodls code doesn’t know “…” | A task goes in quotes: woodls code "fix the failing test" |
| … isn’t a git repository: worktrees need one | git init, then a first commit; --bg and --worktree need git |
| go mode needs a sandbox for commands | Use --mode ask: this computer has no sandbox woodls can use |
| A card that won’t answer | Its keys wake 400 ms after it shows; fast typing goes to your message instead |
Anything else: /feedback WORDS inside woodls code writes a report on this computer for you to read
and send.
Why it isn’t serving
In its own words.
woodls host --status says what it’s doing now, and why. The host view’s states are explained in
Lend your computer.
| It says | What to do |
|---|---|
| woodls host needs a model’s file the first time | Run woodls host in a terminal and say yes to the download, or woodls models get first |
| this computer can’t host for the public network yet: its smallest model, gpt-oss 20B, needs … | The Mac has too little memory for a public model. It can lend Qwen3 8B to your private groups |
| woodls’ public network takes only computers on … for now, and this one runs on … | PCs lend to private groups for now: woodls host --no-public --group NAME |
| … doesn’t fit here | woodls models says which fit; on a Mac, --memory can give models more |
| hosting needs macOS 15 or later | Update macOS, in System Settings, Software Update |
| asleep: another app is using the GPU; the model is released | Nothing: it comes back once the app has been quiet for 5 minutes |
| paused: this computer is on battery | Plug it in, or it waits |
| woodls’ server could not admit this host now | The reason follows. A slow answer or a different one means this computer can’t serve that model for the public network now |
| woodls’ server doesn’t list it now: woodls doctor checks the way in | Run woodls doctor |
| woodls’ engine runs only in its sandbox, and it can’t be made here | See the sandbox on Ubuntu 24.04 |
| esc leave: hosting ends with Ubuntu’s last window | On Windows, Windows couldn’t hold lending: keep an Ubuntu window open, and send us ~/.woodls/host.log |
A computer earns only when someone asks, so a quiet night can pass with no jobs. woodls public
shows the network now, and woodls status what this computer earned today.
Between your computers
Directly or through a relay.
| It says | What to do |
|---|---|
| no computer in private group home serves a model right now | A computer of the group must run woodls host --group home; woodls group members home shows which are online |
| in that private group already, as home | Nothing: this computer is in it |
| the computer the link names wasn’t heard on this network | It isn’t running woodls host or the invite. With relays off, both computers must be on one network |
| this computer is in another group called home already, so this one goes by home-2 here | Nothing: use home-2 on this computer, or join with --as NAME |
| friends in … won’t be served this model file: it isn’t on woodls’ list | A friend’s computer is served gpt-oss 20B, Qwen3.6 35B or Qwen3 8B: lend one of those |
| didn’t answer: it’s asleep, off, or not running woodls host | Ask its owner another way |
On Windows, six digits are often not heard through WSL2: use the link or the spoken code. If the group’s owner computer has been off for longer than its list lasts, the group waits for it to come back.
Your choice
woodls never changes it.
Ubuntu 24.04 (and 23.10 on) stops programs making the user namespaces bubblewrap’s sandbox needs,
through AppArmor. There, woodls code asks before every command, go ahead is off, woodls’ engine
can’t start, and lending doesn’t start. woodls doctor says so, with bubblewrap’s own words:
“loopback: Failed RTM_NEWADDR: Operation not permitted”.
If you choose to allow it, the narrow way lets bubblewrap alone make them. Save this as the file
/etc/apparmor.d/bwrap:
abi <abi/4.0>,
include <tunables/global>
profile bwrap /usr/bin/bwrap flags=(unconfined) {
userns,
include if exists <local/bwrap>
}Then load it, and check:
sudo apparmor_parser -r /etc/apparmor.d/bwrap
woodls doctorTo take it away, run this, then delete the file:
sudo apparmor_parser -R /etc/apparmor.d/bwrapWith the profile, any program can use bubblewrap to make a sandbox, which is what bubblewrap is for. This was tried on Ubuntu 24.04.5 on 2 November 2026. In WSL2, Ubuntu runs on Microsoft’s kernel, and the installer checks there whether the sandbox works.
A person
The founder reads every message.
Write to hello@woodls.com, or reply to your invite email. Say what you
ran, what it said, and which computer it is. If woodls doctor printed something, paste it in,
after reading it for anything you’d rather not send. For a security problem, put “security” in the
subject. Help has the rest.