Documentation
Enscrive gives a project — and any agent working in it — a private, durable memory it can write to, search, and correct.
Self-hosted local mode is the supported path today. You run
the whole stack on one machine with enscrive start; nothing
leaves your box except the embedding calls you configure. The managed API at
api.enscrive.io is pre-launch and not yet serving requests — the
quickstart takes the local path end to end.
New here? Go straight to Install, then the quickstart. If you are an AI agent, or you are setting one up, read For agents.
Getting started
Install the CLI, stand the stack up on your own machine, and give a project a memory.
- Install — Install the enscrive CLI with one command, and what the installer verifies before it puts a binary on your machine.
- Quickstart — Give a project its own durable memory on your own machine — install, start the stack, then remember, recall, and retire.
- For agents — How an AI agent uses Enscrive as per-project memory — the self-teaching AGENT.md, the JSON envelope, and the endpoint-to-command map.
Reference
Every command and flag, generated from the binary itself.
- CLI reference — Every enscrive command and flag, generated from the binary's own --help.
Platform
What Enscrive commits to, the core concepts, and how the API behaves.
- Principles — The commitments that shape how Enscrive is built and what you can count on as a customer.
- Concepts — The three core concepts of the Enscrive platform — Environments, Corpora, Voices — and how they work together.
- Authentication — API keys, headers, and the scope of a single key.
- Errors — HTTP status codes returned by /v1, what they mean, and how to recover.
API reference
The /v1 HTTP surface, endpoint by endpoint.
- Corpora — Create, list, inspect, update, and delete corpora — the stores of documents and embeddings in Enscrive.
- Voices — Chunking + retrieval configurations. Create, update, compare, and promote voices across environments.
- Search — Neural search over a corpus, with and without voice-tuned retrieval.
- Ingest — Push documents into a corpus. Always-async job-based ingest, fingerprint-based deduplication, prepared (two-phase) ingest.
- Jobs — Async job ledger — list, inspect, cancel, retry, abandon. Shared across ingest, commit, eval campaigns, and corpus delete.
- Batch-sets — First-class entity for tracking and recovering provider batch ingest. State machine, atomic-promote guarantees, and how to inspect them.
- Revisions — Enscrive Revisions — the timeline of point-in-time restore points for your tenant, and the verified restore flow.
- Billing — Subscription tiers, the prepaid wallet, metered usage, soft-stop floors, and the versioned public rate card.
Not documented here yet
These parts of the /v1 surface are real but not yet written up
on this site: document browsing, chunking, segmentation, embeddings, evals,
logs, usage, metrics, rate limits, and the admin surface. Until they land,
the CLI reference covers every command the binary
exposes for them, and enscrive <command> --help is
authoritative.