Introduction

Welcome to the Fisher documentation!

What is Fisher?

Fisher is a .NET library for building applications using a document-oriented database approach and Event Sourcing, backed by SQLite.

Fisher is part of the Critter Stack ecosystem and mirrors the API patterns of Marten (PostgreSQL) and Polecat (SQL Server), so a team already using either one will recognise nearly everything here.

TIP

If you've used Marten or Polecat before, you'll feel at home. Same interface names, same session patterns, same projection model. What differs is what SQLite makes cheap and what it makes impossible — and this documentation says which is which rather than leaving you to find out.

The thing that makes Fisher different from its siblings is not the SQL dialect. It is that there is no database server. SQLite runs inside your process, a store is a file, and the "network round trip" that shapes so much of a document database's design simply is not there.

Fisher is built on:

• JasperFx.Events — the shared event, projection and daemon abstractions the whole Critter Stack implements,
• Weasel.Sqlite — schema management, migrations and data access,
• Weasel.Storage — the dialect-neutral document and event storage runtime extracted from Marten.

Fisher supplies the SQLite dialects and the storage seams. Everything above them is shared with Marten and Polecat, which is why a projection written for one store runs unaltered on another.

Main Features

Feature	Description
Document Storage	Store entities as JSON documents with full LINQ querying support.
Event Store	Full event store with stream management, projections and subscriptions.
Strong Consistency	Documents and events commit in one SQLite transaction.
LINQ Querying	Where, ordering, paging, projections, grouping, aggregates and joins.
Event Projections	Inline, live and asynchronous read models.
Automatic Schema Management	Weasel.Sqlite creates and migrates tables.
Optimistic Concurrency	Guid version or numeric revision.
Multi-Tenancy	Conjoined tables, or a database file per tenant.
Async Daemon	Background projection processing.
ASP.NET Core Integration	Streaming JSON results and ETag handling.
EF Core Integration	A DbContext writing inside Fisher's transaction.

Critter Stack Ecosystem

Library	Purpose
Marten	PostgreSQL document database and event store
Polecat	SQL Server document database and event store
Wolverine	Messaging and command processing
JasperFx	Core framework and event sourcing abstractions
Weasel	Database schema management

Fisher vs Marten and Polecat

The API is deliberately the same. The storage decisions are not, and these are the ones worth knowing up front:

Concern	Marten (PostgreSQL)	Polecat (SQL Server)	Fisher (SQLite)
Schemas	real schemas	real schemas	none — folded into the table prefix
JSON	jsonb	native json	TEXT + json1 functions
Timestamps	timestamptz	datetimeoffset	ISO-8601 TEXT, fixed width, UTC
Booleans	boolean	bit	INTEGER 0/1
Guids	native uuid	uniqueidentifier	lowercase canonical TEXT
Event sequence	sequence	IDENTITY	INTEGER PRIMARY KEY AUTOINCREMENT
Document upsert	INSERT … ON CONFLICT	MERGE	INSERT … ON CONFLICT … RETURNING
Append concurrency	advisory lock	UPDLOCK, HOLDLOCK	BEGIN IMMEDIATE
Writers	many	many	one per database file
Serialization	STJ or Newtonsoft	System.Text.Json	System.Text.Json

The single-writer row is the one that propagates. It is why Fisher's unit of work is strictly sequential, why transaction participants and raw SQL commands matter more here than on either sibling, why a busy retry is a real policy rather than a formality, and why database-per-tenant is a performance feature and not only an isolation one.

What Fisher deliberately does not do

• There is no message bus. Projections can publish side effects, and the default outbox drops every message. Delivery is a bus integration's job here as it is on Marten and Polecat.
• There is no table partitioning. SQLite has no partition functions or schemes, and the nearest equivalent carries none of the operational properties that make the feature worth having.
• There is no hot-cold daemon failover. Leader election across nodes means several processes sharing one file, which SQLite does not make safe. The daemon runs Solo.
• Fisher ships no binary event serializer. The seam exists; choosing MessagePack or protobuf is your application's decision about how its data ages.

History and Origins

The Critter Stack names its projects after animals, and the mustelid family has been good to it. A fisher is a small, quick North American mustelid — a cousin of the marten and the polecat, and the smallest of the three, which is about right for the store that fits in a file.