Dash documentation

Troubleshooting

Verified fixes for the most common issues.

Command not recognized

dash: command not found after installing:

Open a new terminal so PATH changes take effect.
Check the global npm bin directory: npm prefix -g. Ensure that directory is on PATH.
On Windows with nvm-windows, PATH points at your active Node installation; switching versions moves the bin path.

PowerShell

npm prefix -g
# add the printed directory to PATH, then reopen the terminal

Installation failures

Engine error about Node. The package requires Node >= 20: node -v, upgrade if needed, retry.
EACCES / permission errors. Run from an elevated prompt or change the npm prefix to a user-writable location.
Slow or blocked downloads. The package is large (~100 MB unpacked, native executable included); confirm your registry and proxy allow the tarball.
Install succeeded but app will not start. Check anti-virus policies — the package runs a bundled native executable.

Invalid API key

"Set API key first with /providers" — no key configured yet. Run /providers and paste a key.
Prompt rejects the key format. NVIDIA NIM keys start with nvapi_; check for copied whitespace.
Key stopped working. Rotate it in the NVIDIA portal, then update via /providers.
Config edited manually and corrupted. Delete ~/.dash/config.json; Dash recreates the default structure on next save, then set the key again.

Unsupported providers or models

Other providers (OpenAI, Anthropic, Google, …) are not integrated yet — keys for them will not be read.
/models shows the models your key can actually reach. If the live list fails, Dash falls back to the built-in Nemotron defaults.
A model id that has left the provider catalog can be re-selected any time with /models.

Network and provider errors

Timeouts / TLS failures against https://integrate.api.nvidia.com: test connectivity and corporate proxies.
401 responses: the stored key is rejected — set a new one.
429 responses: rate limits on your NVIDIA account; retry later or lower request frequency.
Web tools fail without model errors: web_search/web_fetch need internet access but not the provider key.

File access or permission problems

File not found although it exists: dash runs from its launch directory — start dash in your project root.
Paths outside the workspace keep asking for approval: expected; that is the safety model. Edit inside the workspace, or approve the specific request.
Nothing changes when asked to search a folder: the folder is inside the fixed exclusion list (.git, node_modules, .dash, dist, build).
Checkpoints missing: snapshots live under the project at .dash/checkpoints/ — do not delete the directory while a revert is pending.

Command not recognized

dash: command not found after installing:

Open a new terminal so PATH changes take effect.

Check the global npm bin directory: npm prefix -g. Ensure that directory is on PATH.

On Windows with nvm-windows, PATH points at your active Node installation; switching versions moves the bin path.

PowerShell

npm prefix -g
# add the printed directory to PATH, then reopen the terminal

Installation failures

Engine error about Node. The package requires Node >= 20: node -v, upgrade if needed, retry.

EACCES / permission errors. Run from an elevated prompt or change the npm prefix to a user-writable location.

Slow or blocked downloads. The package is large (~100 MB unpacked, native executable included); confirm your registry and proxy allow the tarball.

Install succeeded but app will not start. Check anti-virus policies — the package runs a bundled native executable.

Invalid API key

"Set API key first with /providers" — no key configured yet. Run /providers and paste a key.

Prompt rejects the key format. NVIDIA NIM keys start with nvapi_; check for copied whitespace.

Key stopped working. Rotate it in the NVIDIA portal, then update via /providers.

Config edited manually and corrupted. Delete ~/.dash/config.json; Dash recreates the default structure on next save, then set the key again.

Unsupported providers or models

Other providers (OpenAI, Anthropic, Google, …) are not integrated yet — keys for them will not be read.

/models shows the models your key can actually reach. If the live list fails, Dash falls back to the built-in Nemotron defaults.

A model id that has left the provider catalog can be re-selected any time with /models.

Network and provider errors

Timeouts / TLS failures against https://integrate.api.nvidia.com: test connectivity and corporate proxies.

401 responses: the stored key is rejected — set a new one.

429 responses: rate limits on your NVIDIA account; retry later or lower request frequency.

Web tools fail without model errors: web_search/web_fetch need internet access but not the provider key.

File access or permission problems

File not found although it exists: dash runs from its launch directory — start dash in your project root.

Paths outside the workspace keep asking for approval: expected; that is the safety model. Edit inside the workspace, or approve the specific request.

Nothing changes when asked to search a folder: the folder is inside the fixed exclusion list (.git, node_modules, .dash, dist, build).

Checkpoints missing: snapshots live under the project at .dash/checkpoints/ — do not delete the directory while a revert is pending.