Add Grok STT as the default transcription provider - #130
Conversation
Grok STT already ships word-level timestamps, diarization, and filler-word retention. Normalize its words into the existing Scribe-shaped schema so pack/render/timeline_view stay unchanged. ElevenLabs remains available via --provider elevenlabs. Vision composites stay local (timeline_view PNGs) — no API to swap.
Existing ELEVENLABS_API_KEY-only .env files keep working with no flags. Either key is enough; if both are set, Grok STT is used unless --provider elevenlabs. Install upserts one variable and never truncates .env.
![cubic-dev-ai[bot]](https://avatars.githubusercontent.com/in/1082092?s=60&v=4){kind=small}
There was a problem hiding this comment.
1 issue found across 5 files (changes from recent commits).
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="install.md">
<violation number="1" location="install.md:124">
P2: The new upsert path runs the user-pasted key through sed as the literal replacement text: `sed "s|^${KEY_NAME}=.*|${KEY_NAME}=${KEY}|"`. In the sed replacement, `&` expands to the entire matched text and `\` is an escape, and the chosen delimiter `|` collides if the key contains one. A key containing any of those characters (e.g. an ElevenLabs/xAI key with `&` or `|`) is silently written corrupted to `.env`, and transcription then fails against the bad key. Escape the key for the replacement context (e.g. `sed "s|^${KEY_NAME}=.*|${KEY_NAME}=${KEY//&/\\&}|"`) or write the line with awk instead so arbitrary key bytes are preserved.</violation>
</file>
Reply with feedback, questions, or to request a fix.
Fix all with cubic | Re-trigger cubic
| # KEY_NAME is XAI_API_KEY or ELEVENLABS_API_KEY | ||
| if grep -q "^${KEY_NAME}=" "$ENV"; then | ||
| tmp=$(mktemp) | ||
| sed "s|^${KEY_NAME}=.*|${KEY_NAME}=${KEY}|" "$ENV" > "$tmp" && mv "$tmp" "$ENV" |
There was a problem hiding this comment.
P2: The new upsert path runs the user-pasted key through sed as the literal replacement text: sed "s|^${KEY_NAME}=.*|${KEY_NAME}=${KEY}|". In the sed replacement, & expands to the entire matched text and \ is an escape, and the chosen delimiter | collides if the key contains one. A key containing any of those characters (e.g. an ElevenLabs/xAI key with & or |) is silently written corrupted to .env, and transcription then fails against the bad key. Escape the key for the replacement context (e.g. sed "s|^${KEY_NAME}=.*|${KEY_NAME}=${KEY//&/\\&}|") or write the line with awk instead so arbitrary key bytes are preserved.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At install.md, line 124:
<comment>The new upsert path runs the user-pasted key through sed as the literal replacement text: `sed "s|^${KEY_NAME}=.*|${KEY_NAME}=${KEY}|"`. In the sed replacement, `&` expands to the entire matched text and `\` is an escape, and the chosen delimiter `|` collides if the key contains one. A key containing any of those characters (e.g. an ElevenLabs/xAI key with `&` or `|`) is silently written corrupted to `.env`, and transcription then fails against the bad key. Escape the key for the replacement context (e.g. `sed "s|^${KEY_NAME}=.*|${KEY_NAME}=${KEY//&/\\&}|"`) or write the line with awk instead so arbitrary key bytes are preserved.</comment>
<file context>
@@ -96,40 +96,52 @@ Figure out which agent you are running under, and register once. A symlink of th
+ # KEY_NAME is XAI_API_KEY or ELEVENLABS_API_KEY
+ if grep -q "^${KEY_NAME}=" "$ENV"; then
+ tmp=$(mktemp)
+ sed "s|^${KEY_NAME}=.*|${KEY_NAME}=${KEY}|" "$ENV" > "$tmp" && mv "$tmp" "$ENV"
+ else
+ printf '%s=%s\n' "$KEY_NAME" "$KEY" >> "$ENV"
</file context>
There was a problem hiding this comment.
3 issues found and verified against the latest diff
Prompt for AI agents (unresolved issues)
Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.
<file name="helpers/transcribe_batch.py">
<violation number="1" location="helpers/transcribe_batch.py:50">
P2: When a transcript already exists, `--provider` is ignored because the batch cache is provider-agnostic. Running `--provider grok` after an ElevenLabs transcript silently reuses the ElevenLabs result; track the provider in the cache decision or invalidate the cached file when it differs from the requested backend.</violation>
</file>
<file name="helpers/transcribe.py">
<violation number="1" location="helpers/transcribe.py:79">
P2: When a caller supplies an ElevenLabs `api_key` but omits `provider`, this ambient-key lookup can select Grok and send the wrong credential, causing authentication failures. Preserve the existing direct-call behavior or require and validate the provider alongside the key.</violation>
<violation number="2" location="helpers/transcribe.py:79">
P2: Adding a second provider makes the existing filename-only transcript cache return stale cross-provider data. A user who previously transcribed with ElevenLabs and now runs the new grok default will get their old Scribe transcripts back (and vice versa) because the cache checks only that `<video_stem>.json` exists. Make the cache provider-aware: include the provider in the cached filename or verify the stored `provider` field on the cache hit, and skip/re-transcribe when it differs.</violation>
</file>
Reply with feedback, questions, or to request a fix.
Fix all with cubic | Re-trigger cubic
| ) | ||
| ap.add_argument("--workers", type=int, default=4, help="Parallel workers (default: 4)") | ||
| ap.add_argument( | ||
| "--provider", |
There was a problem hiding this comment.
P2: When a transcript already exists, --provider is ignored because the batch cache is provider-agnostic. Running --provider grok after an ElevenLabs transcript silently reuses the ElevenLabs result; track the provider in the cache decision or invalidate the cached file when it differs from the requested backend.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At helpers/transcribe_batch.py, line 50:
<comment>When a transcript already exists, `--provider` is ignored because the batch cache is provider-agnostic. Running `--provider grok` after an ElevenLabs transcript silently reuses the ElevenLabs result; track the provider in the cache decision or invalidate the cached file when it differs from the requested backend.</comment>
<file context>
@@ -44,6 +46,12 @@ def main() -> None:
)
ap.add_argument("--workers", type=int, default=4, help="Parallel workers (default: 4)")
+ ap.add_argument(
+ "--provider",
+ choices=PROVIDERS,
+ default=None,
</file context>
|
|
||
|
|
||
| def load_api_key(provider: str | None = None) -> str: | ||
| provider = resolve_provider(provider) |
There was a problem hiding this comment.
P2: When a caller supplies an ElevenLabs api_key but omits provider, this ambient-key lookup can select Grok and send the wrong credential, causing authentication failures. Preserve the existing direct-call behavior or require and validate the provider alongside the key.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At helpers/transcribe.py, line 75:
<comment>When a caller supplies an ElevenLabs `api_key` but omits `provider`, this ambient-key lookup can select Grok and send the wrong credential, causing authentication failures. Preserve the existing direct-call behavior or require and validate the provider alongside the key.</comment>
<file context>
@@ -27,22 +30,53 @@
+
+
+def load_api_key(provider: str | None = None) -> str:
+ provider = resolve_provider(provider)
+ name = KEY_FOR_PROVIDER[provider]
+ v = _read_key(name)
</file context>
|
|
||
|
|
||
| def load_api_key(provider: str | None = None) -> str: | ||
| provider = resolve_provider(provider) |
There was a problem hiding this comment.
P2: Adding a second provider makes the existing filename-only transcript cache return stale cross-provider data. A user who previously transcribed with ElevenLabs and now runs the new grok default will get their old Scribe transcripts back (and vice versa) because the cache checks only that <video_stem>.json exists. Make the cache provider-aware: include the provider in the cached filename or verify the stored provider field on the cache hit, and skip/re-transcribe when it differs.
Prompt for AI agents
Check if this issue is valid — if so, understand the root cause and fix it. At helpers/transcribe.py, line 75:
<comment>Adding a second provider makes the existing filename-only transcript cache return stale cross-provider data. A user who previously transcribed with ElevenLabs and now runs the new grok default will get their old Scribe transcripts back (and vice versa) because the cache checks only that `<video_stem>.json` exists. Make the cache provider-aware: include the provider in the cached filename or verify the stored `provider` field on the cache hit, and skip/re-transcribe when it differs.</comment>
<file context>
@@ -27,22 +30,53 @@
+
+
+def load_api_key(provider: str | None = None) -> str:
+ provider = resolve_provider(provider)
+ name = KEY_FOR_PROVIDER[provider]
+ v = _read_key(name)
</file context>
1 participant
Why
video-use currently hard-depends on ElevenLabs Scribe. Grok STT already has the three things the cut pipeline needs: word-level timestamps, speaker diarization, and filler-word retention.
Vision does not need a new API.
timeline_view.pystill emits local filmstrip/waveform PNGs; the host agent (Grok, Claude, Codex, …) reads them with whatever vision it already has.What
helpers/transcribe.pyaccepts eitherXAI_API_KEYorELEVENLABS_API_KEY..env(today's install) keeps working with no flags.--provider elevenlabs.type/speaker_id/ synthesizedspacingtokens) so pack/render/timeline_view stay unchanged.install.mdupserts one key line and never truncates.env.Not in this PR