The usual failure points
T3 Code, like other coding agents, tends to fail in a few predictable places: install, authentication, provider configuration, model selection, and the build step. Work through these in order.
Install fails
Symptom: the installer stops or the binary is missing.
Fix: check the OS and architecture. On Linux use the shell installer, on Windows use PowerShell. Confirm the download completed and that the binary is on your PATH. If the connection dropped, rerun the installer. Fikra Code resumes downloads, so a retry continues instead of starting from zero.
Authentication errors
Symptom: 401 or "invalid key".
Fix: confirm you are using the right key type. A Fikra Code key is for Fikra Code. A normal API key keeps the usual API catalog. Paste the key at fikra login or set FIKRA_API_KEY.
Provider configuration
Symptom: the agent ignores your provider or hits the wrong host.
Fix: set the base URL explicitly to https://api.fikraapi.co.ke/v1 and confirm the provider is OpenAI-compatible. Check that no stale config overrides the base URL.
Model not available
Symptom: model errors or empty responses.
Fix: use an id from the live catalog at https://fikraapi.co.ke/code/models. Only ids marked In API are currently callable.
Build step fails
Symptom: compile or packaging errors.
Fix: read the first error, not the last. Most build failures are a missing toolchain or a dependency mismatch. On Windows, enable git long paths before fetching dependencies. On Linux, install the Rust toolchain named by the project.
Network stalls
Symptom: long pauses, then a timeout.
Fix: retry with streaming. Avoid large context requests on a metered connection. Move to a model that fits your bandwidth.
Updates break the install
Symptom: the agent works, then fails after an update.
Fix: keep a known good version, and use the official installer. Fikra Code stores versions under ~/.fikra/packages and keeps the two newest.
Get a clean start
- Create a Fikra Code key at https://fikraapi.co.ke/dashboard/code
- Install on Linux or Windows.
- Run
fikra login. - Start with the free DeepSeek model.
Docs: https://fikraapi.co.ke/code/docs