[What landed in August](#what-landed-in-august) -
[Effect v4 Release Candidate](#effect-v4-release-candidate) -
[Zero External Dependencies in Core](#zero-external-dependencies-in-core) -
[Native PostgreSQL Client](#native-postgresql-client) -
[Pull-based Sockets](#pull-based-sockets) -
[Native Schema Arbitraries](#native-schema-arbitraries) -
[Schema](#schema) -
[Match](#match) -
[RPC](#rpc) -
[Tracing](#tracing) -
[AI & Streaming Tools](#ai--streaming-tools) -
[MCP](#mcp) -
[Cluster & Durable Execution](#cluster--durable-execution) -
[SQL](#sql) -
[Platform & Runtime](#platform--runtime) -
[Effect Core](#effect-core) -
[Schedule](#schedule) -
[Graph](#graph) -
[CLI](#cli) -
[Optics](#optics) -
[Tooling](#tooling) -
[Documentation](#documentation) -
[Keep up with the Effect v4 RC](#keep-up-with-the-effect-v4-rc)
August was the month Effect v4 crossed the final threshold before stable.
On August 12th, we announced the Effect v4 Release Candidate, and the Effect v4 core package now has zero external dependencies.
What landed in August
Effect v4 Release Candidate
After months of beta releases, real-world usage, and a deep audit of the entire codebase, Effect is ready for 4. No broad breaking changes are planned for the RC cycle, any change that does land will be narrow in scope, clearly communicated, and accompanied by a migration path.
The RC is an invitation to migrate real applications and validate against production workloads. If you find anything that should stop the stable release, we want to know now.
Read the full RC announcement for context, migration guidance, and what comes next.
Zero External Dependencies in Core
The effect
core package has no external dependencies. This is the result of a sustained effort throughout the RC cycle to internalize or eliminate third-party dependencies, including replacing pg
with a native PgProtocol
client, removing MessagePack and the MIME dependency from @effect/platform-node
, and migrating NodeRedis
from ioredis
to redis
(node-redis). Effect already replaces 50+ common packages (observability, HTTP, CLI, database drivers, dependency injection), and doing so with no external surface area of its own makes it uniquely resilient to supply chain attacks.
Native PostgreSQL Client
Replaced the pg
dependency in @effect/sql-pg
with a native PgProtocol
client, reducing external dependencies and giving the team full control over the connection lifecycle. PGlite listen
behavior was also aligned with standard PostgreSQL semantics.
Pull-based Sockets
The Socket
API was reworked around a pull-based reader with end-to-end backpressure:
-
reconnects are now as simple as
Effect.retry -
the WebSocket client interface is unified across runtimes (Node, Bun, Deno, browser) with typed handshake headers
-
fixed the Node platform writing HTTP responses onto already-upgraded WebSocket connections.
Native Schema Arbitraries
Schema’s Arbitrary generation and shrinking were reworked with a native implementation, replacing the previous approach for more predictable and efficient property-based test data generation, with no external dependency.
Schema
New APIs - added partial matching to Schema.TaggedUnion
, configurable schema reference policies, and a protocol adapter for the latest MCP specification revision.
Performance - optimized synchronous schema parsing. Fixed projection of checks and annotations through artificial JSON encodings. Fixed Schema
class equivalence derivation. Match
Added Match.fn
selector matchers for more ergonomic pattern matching without boilerplate.
RPC
- Made server notifications first-class RPC messages.
- Added HTTP stream backpressure to prevent unbounded memory growth under slow consumers.
Tracing
- Improved tracing performance in span creation and HTTP tracer middleware.
- Shared tracer hex ID generation across spans.
- Documented the public random hex contract.
- Fixed log annotations overwriting active span correlation identifiers.
- Fixed disabled tracer timing still recording span end times.
AI & Streaming Tools
Streaming tool handlers are now executed eagerly, supporting interruption and synthesized results instead of buffering the full response. Other fixes and improvements:
- preserved encoded AI tool parameters for manual resolution
- fixed
Deferred
clearing resumes before resuming waiters on completion - corrected max output tokens for Claude Opus 4.6, 4.7, 4.8 and Sonnet 4.6
- fixed OpenRouter emitting cumulative tool-parameter prefixes as deltas
- fixed OpenRouter reasoning-end metadata casing
- fixed OpenRouter streamed citation end offsets overwriting start offsets
- fixed OpenAI telemetry response attributes using the request namespace type
- fixed batched encrypted event records reusing per-record randomization metadata
- fixed unknown MCP cancellation suppressing a later reused request ID.
MCP
- Fixed MCP conformance issues and added a test suite.
- Emitted MCP tool output schemas.
- Added a protocol adapter for the latest MCP specification revision.
Cluster & Durable Execution
- Fixed a cluster shutdown hang from abandoned requests.
- Fixed defecting execution-plan observers replacing outcomes and leaving events unpaired.
- Fixed
TxPubSub
subscriber release interrupting after hub shutdown. - Fixed SQL
findById
resolvers invokingNonEmpty
callbacks with empty batches. - Fixed SQL resolvers executing empty batches.
- Used migrations for SQL persisted queues.
- Added integration coverage for entity shutdown, replay, workflow owner-loss, and cluster transport failure scenarios.
- Bounded teardown interrupt classification.
- Ensured transient routing states are never surfaced for persisted messages.
- Defaulted TCP cluster serialization to
SchemaBinary
.
SQL
- Added D1 batch support.
- Ran SQL persistence cleanup in bounded batches to avoid lock contention.
- Fixed
TMap.remove
/removeAll
clearing entire buckets on hash collision. - Fixed readonly SQLite transactions.
- Propagated failed transaction begin as a typed error.
- Fixed
NodeStream
zeromaxBytes
handling.
Platform & Runtime
- Fixed Bun and Deno
fileWebResponse
ignoring range options. - Fixed closing an older Bun server not properly tearing down.
- Fixed client-server-client
FormData
round-trip. - Fixed
rcRef
leaking resources acquired before acquisition failure. - Fixed
scopedRef.set
orphaning a replacement when old cleanup defects. - Fixed buffered worker send failures escaping the
WorkerError
channel. - Fixed
fiber.joinAll
erasing input fiber error types. - Fixed
NodeTerminal
readline staying alive between prompts. - Restored
Effect.timeout
error message. - Aligned type IDs with module paths.
- Normalized collection count handling.
Effect Core
- Restored
Effect.head
, which had been inadvertently removed. - Merged resource usage and finalizer failures.
- Preserved
Context
accessor with loose object spread. - Fixed
Effect.fromOption
inline inference. - Normalized unbounded
PubSub
replay capacities. - Respected zero Node socket open timeouts.
Schedule
Added support for refinements in Schedule.while
.
Graph
Fixed edge mutability and A* heuristics.
CLI
- Added context-based CLI prompt themes for more customizable interactive prompts.
- Fixed quotes escaping in generated shell completion descriptions.
Optics
Added standalone dual Optic
functions.
Tooling
Effect LSP rules are now available as custom, type-aware OXLint rules, with support for vite-plus. Effect’s API reference documentation for both v3 and v4 is now available directly on the Effect website, with an improved search experience.
A skill for Effect v3 to v4 migration is now available on skills.sh, enabling near-fully automated migrations with a coding agent. Documentation
- Recommended
Effect.fnUntraced
in the AI guide. - Documented
Match
and RPC migration changes for teams upgrading from v3. - Added
SQL
,HttpApi
testing, and CLI schema coverage to the AI docs. - Fixed
KeyValueStore
file backend docs.
Keep up with the Effect v4 RC
If you missed the earlier updates, the [July recap](https://effect.website/blog/effect-v4beta-july-recap/), the [June recap](https://effect.website/blog/effect-v4beta-june-recap/), and the [February–May recap](https://effect.website/blog/effect-v4beta-launch-to-may-recap/) cover the full beta period.
For week-by-week updates, follow the [This Week in Effect](https://www.effect.website/blog?category=this-week-in-effect#blog-grid) series.
To try the Effect v4 RC:
npm install effect@rc
pnpm add effect@rc
yarn add effect@rc
bun add effect@rc
Happy Effecting. 🚀