Skip to content

Troubleshooting Echo ​

Start with the smallest reversible check. Do not delete models, History, workflows, or configuration until the corresponding Events and Logs have been inspected.

Recovery order ​

  1. Open Settings → Diagnostics → About and note the Echo version, platform, and distribution channel.
  2. Reproduce the issue once with the smallest input or workflow.
  3. Read the newest Warning or Error in Events.
  4. Search Logs for the same timestamp, model, workflow, or subsystem.
  5. Correct one setting, permission, or input and retry once.
  6. Escalate with the evidence described under Diagnostics.

Permission failures ​

For an unavailable microphone, select the intended input device under General, run the microphone test, and confirm operating-system microphone permission for the current Echo build.

On macOS, Auto paste and some direct-build input features require Accessibility permission. Use the warning in General to open System Settings. Text should remain on the clipboard when Auto paste cannot press the paste shortcut, so paste it manually while diagnosing access.

If a global shortcut does not register, choose a combination not reserved by another application. Mac App Store builds accept keyboard recording shortcuts and reject mouse-button recording bindings.

Microphone and recording issues ​

  • Confirm the input device in General and test it there.
  • Check that another application has not taken exclusive control of the device.
  • Confirm the selected transcription model is downloaded and active.
  • For always-on, look for the explicit no-model error and review the configured trigger phrase.
  • If recording starts but output is empty, inspect Events before changing segmentation parameters.

Model and download issues ​

Model acquisition needs network access, disk space, and a complete download. Retry the row's download action, wait for progress to finish, then select the model. A file on disk is not proof that activation and engine rebuild succeeded.

If activation still fails, record the model identifier shown in Events and the first matching error in Logs. Avoid publishing a fixed model list in a support report; use the identifier shown by the current catalogue.

Provider issues ​

  • Cloud model: verify sign-in, quota in About, network access, and that Privacy mode is not enabled.
  • Detected local model: start Ollama or LM Studio, load a model, choose Rescan, then select it.
  • Custom endpoint: verify URL, provider type, model identifier, and required credential with the built-in test. Do not paste the credential into Events, Logs, or a report.
  • Unexpected destination: stop the run and re-check the selected profile plus every HTTP, download, plugin, and nested recipe action.

Workflow and run issues ​

Open the workflow in the editor and use its test path with minimal input. Check required variables, action configuration, connections, terminal states, and side-effect permissions. Then inspect Workflow Runs for the failing step and compare it with the Events timestamp.

For a shared recipe, review permissions recursively before installing or running it. For Shell, file, application, keyboard, HTTP, and plugin actions, confirm the operating-system and trust boundary separately.

Escalating to support ​

Open About → Report a Bug. Include:

  • a short title and exact reproduction steps;
  • expected and actual results;
  • app version, platform, and direct/store distribution;
  • the first relevant error and affected model or workflow name;
  • only the shortest necessary log excerpt.

The form can attach recent logs. Uncheck Include application logs if you cannot safely share them. Remove transcription content, paths, tokens, account data, endpoint details, and unrelated lines according to Safe redaction.

Released under the MIT License.