Fliki
AIProduce AI voiceovers with Fliki from inside a flow. Agents browse languages, dialects, voices, and styles, generate audio synchronously or as background tasks, poll generation status, and check account usage.
What This Integration Enables
Fliki turns text or SSML into natural speech across a catalog of thousands of voices in dozens of languages, and the connector fixes the part that breaks automations: Fliki hosts generated audio for only 30 days, so FlowRunner downloads every result and saves it to durable file storage, returning a permanent URL alongside the temporary one. That one behavior is the difference between a voiceover pipeline and a folder of dead links. Agents browse languages, dialects, voices, and speaking styles from the live catalog, generate audio synchronously or as background tasks for large batches, and watch quota so a big job never dies halfway. Whether a voice reaches customers is a [human-in-the-loop](/concepts/human-in-the-loop) decision, because ears catch what text review cannot. - Voice articles, product descriptions, and newsletters in a consistent brand voice - Localize audio per market with language, dialect, and voice selected per flow - Run large batches asynchronously with custom polling, without blocking the flow - Check remaining quota before committing a batch against the allowance - Control pacing and emphasis with SSML where the default read is not the right read
Without FlowRunner
With FlowRunner
Use Case Scenarios
The blog that publishes in two formats
Every post that goes live in [WordPress](/integrations/wordpress) triggers the narration flow: segment the text, apply the house SSML conventions for pacing, voice it with Generate Audio, and attach the permanent file URL to the post as an audio player. The archive copy lands in [Google Drive](/integrations/google-drive-api) and the editor gets a [Slack](/integrations/slack) note with the listen link. Readers who commute become listeners, and the marginal cost of the audio edition drops to a review listen. The permanent URL matters here: the player embedded in a post from last spring still resolves, because the file lives in FlowRunner storage rather than on a link with an expiry date.
One announcement, five markets
A product announcement needs to sound native in each market. The agent walks List Languages and List Dialects to pin the right accent per region, then List Voices to pull each market's approved voice id and speaking style. Generate Audio produces the set, and each file routes to that market's reviewer, a native speaker, who listens before anything publishes. The localization step that used to gate every launch becomes a parameter sweep with a human ear at the end, and the voice ids live in the flow's configuration where the next launch reuses them.
Batch narration without a blocked flow
A back-catalog project needs hundreds of product descriptions voiced. Blocking a flow on each generation would take all day, so the agent checks Get Usage, then submits jobs with Create Audio Task and collects job ids. A polling loop calls Get Generation Status, downloads completed audio promptly to beat the 30-day hosting window, and logs each permanent URL against its SKU in [Airtable](/integrations/airtable). Failures retry with backoff, and the batch report tells a human exactly what completed, what failed, and what still needs a decision.
Human-in-Loop Highlight
Generate Audio draws down a finite character allowance with every call, which is why Get Usage exists and why big batches get a quota check and a human sign-off before submission: a mis-scoped batch spends an allowance that does not come back. The costlier failure is the one that reaches ears. A voiceover that mispronounces the founder's name, reads a price wrong, or hits the wrong tone in a sensitive announcement carries the brand's voice literally, and audio errors are harder to catch in review queues because someone must actually listen. So nothing generated here auto-publishes to a podcast feed, a product page, or a phone prompt. The editor's listen-through is the release gate, and the agent's job is to make that listen the only manual step left.
Agent Capabilities
7 actionsAudio Generation
3- Generate Audio Converts text or SSML into speech with a chosen voice, waits for completion, downloads the result, and saves it to FlowRunner file storage, returning a permanent file URL alongside the temporary Fliki one. Content runs to about 2,000 characters per call, with SSML available for pace and emphasis.
- Create Audio Task Submits an asynchronous generation job and returns immediately with a job id. The right call for large batches and non-blocking flows with custom polling logic.
- Get Generation Status Polls an asynchronous job and returns the temporary download URL when complete. Results should be downloaded promptly, since Fliki hosts audio for only 30 days.
Reference
3- List Languages Retrieves every supported language with id, name, and code. The first step in localization flows and the filter key for voices.
- List Dialects Retrieves supported dialects and accents by id, for narrowing a language to American versus British English and the like.
- List Voices Retrieves the voices for a language, optionally narrowed by dialect, each with display name, gender, and supported speaking styles. The catalog read behind a locked-in brand voice.
Account
1- Get Usage Retrieves current billing-period consumption and remaining allowance. The pre-flight check that keeps a large batch from dying mid-run.
Frequently Asked Questions
What can FlowRunner do with Fliki?
FlowRunner agents can run List Languages, List Dialects, and List Voices in Fliki, plus 4 more actions.
Does connecting Fliki to FlowRunner require OAuth?
No. Fliki connects to FlowRunner with an API key, no OAuth flow required.
Can Fliki trigger a FlowRunner workflow automatically?
Fliki doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Fliki
$100 in credits. No card required. Connect in minutes.