Maintenance Mode: The platform is currently paused for upgrades. Some features may be unavailable.
Public Beta: Epist SDK v1.2

A memory layer for spoken knowledge.

Epist turns audio into a searchable, time-aware knowledge base. Reason over conversations, meetings, and voice notes—don't just transcribe them.

epist-temporal-engine
Live Index
Context
Architecture Sync
Answer Derived from Audio

The migration timeline was finalized during the Architecture Sync (14:02). Sarah explicitly requested a two-week buffer for load testing, which David approved at 14:05.

"It was discussed, but where?"

Stop scrubbing through 60-minute recordings to find a 10-second decision. Epist indexes the concept, locating the exact moment instantly.

"I need exact wording, not summaries."

LLM summaries hallucinate or gloss over nuance. Epist provides the verbatim transcript segment anchored to the original audio timestamp.

"How did our thinking evolve?"

Reason over chronology. Epist maps the evolution of a project across weeks of meetings, voice notes, and interviews.

Audio is where decisions hide. Epist makes them retrievable.

Ask real questions of real audio.

No keywords. Just natural language reasoning over your data.

workspace: product_eng_syncsReady
E

The primary concern was the click target size on smaller devices. Design argued that the current 44px hit area causes mis-clicks in the navigation drawer.

Designed for the messy reality of human speech.

Read the research

Audio-Aware Segmentation

Speech doesn't have paragraphs. We use prosody, silence detection, and speaker turns to chunk data semantically before embedding.

Semantic Memory

Meaning survives paraphrasing. Search for "money concerns" and Epist retrieves "budgetary constraints" automatically.

Time-Respecting Retrieval

Answers always come with "when." We maintain a strict chronological graph to understand the order of events.

Three lines to indexed intelligence.

We abstracted the hard parts. No vector DBs. No chunking logic. No RAG plumbing. Just your audio and a question.

UploadSupports mp3, wav, m4a, flac up to 500MB
IndexAsync processing with webhooks
ReasonSub-second query latency
main.py
import epist

client = epist.Client()

# 1. Ingest & Index (Async)
audio = client.upload("./meeting_recording.mp3")
index = client.index.create([audio.id])

# 2. Query with Citational Reasoning
answer = index.query(
    "What was the verdict on the database migration?"
)

print(answer.text)
# >> "The team agreed to migrate to Postgres..."

print(answer.citations)
# >> [{"start": 842.5, "end": 855.2, "confidence": 0.98}]

Built as infrastructure. Used everywhere.

Meeting Archives

Searchable company memory

Podcast Intelligence

Deep discovery & show notes

Research Memory

Synthesize user interviews

Internal Tools

Support call QA & Compliance

Start free. Scale with volume.

No credit card required to get started.

Sandbox
$0/mo
  • 5 hours transcription / mo
  • Standard vector search
  • No credit card required
Starter
$19/mo
  • 20 hours transcription / mo
  • Basic vector search
  • API Access
Best Value
Pro
$49/mo
  • 100 hours transcription / mo
  • High-priority Reranking
  • API Access & Shared Seats

Not everything important is written down.

Epist exists to make spoken knowledge durable, searchable, and usable.