Resili Beta 1

Resili Beta 1

Resili Beta 1 is the first public beta of the composable TypeScript resilience toolkit for Node.js and LLM applications.

This release marks the transition from alpha experimentation to a hardened public API. The Core, HTTP adapters, LLM foundation, and provider adapters have completed API reviews, packaging verification, cancellation hardening, and public-registry validation.

What's Included

Core Resilience

@resili/core@0.2.0-beta.1

Nine composable resilience policies:

• Retry
• Timeout
• Circuit Breaker
• Bulkhead
• Rate Limiter
• Fallback
• Hedge
• Request Deduplication
• Cache

The policy pipeline is deterministic and composable, with support for execution context, deadlines, cancellation, events, and metrics.

HTTP Adapters

• @resili/fetch@0.2.0-beta.1
• @resili/axios@0.2.0-beta.1
• @resili/undici@0.2.0-beta.1

All three adapters support caller-initiated AbortSignal cancellation while preserving Resili timeout and retry behavior.

Adapters also expose lifecycle access through on() and destroy() without exposing the underlying Core client.

LLM Resilience

@resili/llm@0.1.0-beta.1

The provider-neutral LLM resilience layer includes:

• generate()
• Pull-through stream()
• Retry and timeout handling
• Streaming commit-point protection
• Usage accounting
• Caller-defined pricing
• Budget Guard
• Cancellation
• Privacy-safe telemetry
• Provider-neutral errors and events

Streaming retries are allowed only before visible text has been committed to the consumer.

Once the first non-empty text delta has been delivered, Resili will not start another provider generation. A post-commit timeout therefore terminates as a non-retryable LLM timeout instead of risking duplicate output.

LLM Provider Adapters

• @resili/llm-openai@0.1.0-beta.1
• @resili/llm-anthropic@0.1.0-beta.1
• @resili/llm-gemini@0.1.0-beta.1

Provider SDK retry layers are disabled so Resili remains the owner of retry behavior.

Supported integrations:

• OpenAI Chat Completions
• Anthropic Messages
• Google Gemini generateContent / generateContentStream

Install

Core

npm install @resili/core

Fetch

npm install @resili/core @resili/fetch

Axios

npm install @resili/core @resili/axios axios

Undici

npm install @resili/core @resili/undici undici

LLM

npm install @resili/core @resili/llm

OpenAI

npm install @resili/llm-openai openai

Anthropic

npm install @resili/llm-anthropic @anthropic-ai/sdk

Gemini

npm install @resili/llm-gemini @google/genai

Both the latest and beta npm dist-tags currently resolve to Beta 1.

Package Versions

Package	Version
@resili/core	0.2.0-beta.1
@resili/fetch	0.2.0-beta.1
@resili/axios	0.2.0-beta.1
@resili/undici	0.2.0-beta.1
@resili/llm	0.1.0-beta.1
@resili/llm-openai	0.1.0-beta.1
@resili/llm-anthropic	0.1.0-beta.1
@resili/llm-gemini	0.1.0-beta.1

Verification

Beta 1 was validated using packages installed exclusively from the public npm registry.

Release verification includes:

• 641 repository tests
• Node.js 20 and Node.js 22
• ESM and CommonJS consumers
• All 8 public packages
• Public npm dependency graph verification
• Single Core and LLM dependency instances
• HTTP caller cancellation
• Retry and timeout behavior
• LLM unary generation
• LLM pull-through streaming
• Pre-commit streaming retries
• Post-commit timeout protection
• OpenAI, Anthropic, and Gemini adapters
• Budget Guard accounting and cleanup
• Telemetry/privacy checks
• Package-content and workspace-leakage checks

The release-blocking streaming regression test confirms that after visible output is committed, a timeout results in exactly one provider generation with no delayed retry or duplicate text.

Beta Status

This is a public beta.

The primary APIs have completed their beta freeze reviews, but changes may still be made before the first stable release based on real-world usage and feedback.

Known scope boundaries include:

• Process-local budget/state management
• No LLM tools support yet
• No multimodal support yet
• No embeddings support yet
• No OpenAI Responses API support yet

Feedback, bug reports, and real-world usage reports are welcome as Resili moves toward its first stable release.

Resili v0.2.0-alpha.3 — Core Hardening

Resili v0.2.0-alpha.3

This release hardens Resili's core runtime and aligns the HTTP and LLM package lines on @resili/core@0.2.0-alpha.3.

Core Hardening

• cache is now a valid relative policy-order anchor
• RequestStarted and RequestCompleted are emitted once per top-level execution
• stats().totals.retries now tracks additional retry attempts
• Added rate limiter onLimit: "wait" support with maxWaitMs
• Rate limiter waiting is FIFO per key and respects cancellation and timeout

circuit, bulkhead, and rateLimiter policy maps in stats() remain empty in this alpha. Live policy-state aggregation is not included yet.

HTTP Packages

The HTTP adapters have been republished against the hardened core. Adapter behavior itself is unchanged.

• @resili/core@0.2.0-alpha.3
• @resili/fetch@0.2.0-alpha.3
• @resili/axios@0.2.0-alpha.3
• @resili/undici@0.2.0-alpha.3

LLM Compatibility Line

The LLM packages have been republished so the entire public package graph resolves to @resili/core@0.2.0-alpha.3.

• @resili/llm@0.1.0-alpha.2
• @resili/llm-openai@0.1.0-alpha.2
• @resili/llm-anthropic@0.1.0-alpha.2

No LLM provider runtime behavior changed in this compatibility release.

Install

Core:

npm install @resili/core@alpha

Fetch:

npm install @resili/core@alpha @resili/fetch@alpha

LLM + OpenAI:

npm install @resili/core@alpha @resili/llm@alpha @resili/llm-openai@alpha openai

LLM + Anthropic:

npm install @resili/core@alpha @resili/llm@alpha @resili/llm-anthropic@alpha @anthropic-ai/sdk
Verification

The release was verified from a clean external npm consumer using only packages downloaded from the public registry.

ESM imports: passed
CommonJS imports: passed
Single deduplicated @resili/core@0.2.0-alpha.3: passed
Single deduplicated @resili/llm@0.1.0-alpha.2: passed
Core hardening smoke tests: passed
Fetch / Axios / Undici adapter smoke tests: passed
OpenAI adapter regression: passed
Anthropic adapter regression: passed
No workspace:* dependency leakage
Alpha Notice

Resili is still in public alpha. APIs and behavior may evolve before the stable release.

Use the alpha dist-tag to explicitly install the current alpha line.

Resili LLM v0.1.0-alpha.4 — Streaming Timeout Fix

Resili LLM v0.1.0-alpha.4

This alpha fixes an important streaming retry boundary discovered during
public-registry verification of the previous alpha.

Fixed

• Prevent retries after the first non-empty text delta has been delivered
  to the consumer.
• Prevent duplicate output from multiple provider generations when a
  per-attempt timeout occurs after streaming has committed.
• Normalize post-commit stream timeouts as:
  • classification: "timeout"
  • retryable: false
• Preserve retries for failures and timeouts that occur before any visible
  text is delivered.
• Preserve existing unary generate() retry behavior.

Streaming

Pull-through streaming is available through @resili/llm with adapters for:

• OpenAI
• Anthropic
• Google Gemini

The provider SDK retry layers remain disabled so Resili owns retry behavior.

Packages

@resili/llm@0.1.0-alpha.4
@resili/llm-openai@0.1.0-alpha.4
@resili/llm-anthropic@0.1.0-alpha.4
@resili/llm-gemini@0.1.0-alpha.3

Core dependency:

@resili/core@0.2.0-alpha.3

Install

npm install @resili/core@alpha @resili/llm@alpha

OpenAI:

npm install @resili/llm-openai@alpha openai

Anthropic:

npm install @resili/llm-anthropic@alpha @anthropic-ai/sdk

Gemini:

npm install @resili/llm-gemini@alpha @google/genai

Verification

The release was verified from a fresh consumer project using packages
installed exclusively from the public npm registry.

The regression test that reproduced the alpha.3 streaming timeout defect
now confirms:

• one provider generation
• no retry after committed text
• no duplicate text
• correct timeout classification
• pre-commit retries remain functional
• unary retry behavior remains unchanged

598 repository tests and the public-registry verification suite pass.

Status

Alpha release. APIs may still evolve before the stable release.

Resili LLM v0.1.0-alpha.1 — Provider-neutral LLM resilience

Resili LLM v0.1.0-alpha.1

This is the first public alpha of Resili’s LLM line. It sits beside @resili/core@0.2.0-alpha.2; it is not a v0.3 core release.

LLM Resilience Foundation

@resili/llm adds a provider-neutral resilience layer for LLM workloads.

• Provider-neutral LLM contracts
• Normalized token usage
• Injectable pricing and deterministic cost calculation
• Budget Guard with estimated-cost preflight
• Unknown model pricing rejected by default
• Typed LLM lifecycle events
• Low-cardinality metrics
• Resili timeout and retry integration
• AbortSignal propagation
• Process-local budget reservations and accounting

OpenAI Adapter

@resili/llm-openai connects OpenAI Chat Completions to the Resili LLM foundation.

• Caller-owned OpenAI client
• Chat Completions support
• OpenAI SDK retries disabled with maxRetries: 0
• Resili remains the retry engine
• Normalized OpenAI error classification
• Usage normalization
• Cost and Budget Guard integration
• AbortSignal support
• API keys, prompts, and responses are not emitted into Resili telemetry by default

Packages

• @resili/llm@0.1.0-alpha.1
• @resili/llm-openai@0.1.0-alpha.1
• Uses @resili/core@0.2.0-alpha.2

Install

Foundation:

pnpm add @resili/core@alpha @resili/llm@alpha

OpenAI:

pnpm add @resili/core@alpha @resili/llm@alpha @resili/llm-openai@alpha openai

Use the alpha dist-tag to install the current public alpha line.

Alpha limitations

• APIs may still change before stable release
• OpenAI Chat Completions only
• No streaming yet
• No Responses API, tools, vision, or embeddings
• No Anthropic, Gemini, Azure OpenAI, or Bedrock adapters yet
• Default budget accounting is process-local
• maxCostPerRequestUsd is an estimated preflight limit, not a guaranteed hard ceiling on actual spend

Resili v0.2.0-alpha.2 — Intelligent Request Management

Resili v0.2.0-alpha.2

This release completes the v0.2 Intelligent Request Management scope.

Added

• Hedged Requests
• Request Deduplication
• Memory Cache
• Typed lifecycle events and low-cardinality metrics
• Benchmark framework
• Updated README and contributing guide

Improved

• Policy integration coverage
• Cancellation, cleanup, and concurrency hardening
• CI validation for the benchmark workspace
• Package metadata and release validation

Packages

• @resili/core@0.2.0-alpha.2
• @resili/fetch@0.2.0-alpha.2
• @resili/axios@0.2.0-alpha.2
• @resili/undici@0.2.0-alpha.2

Status

This is an alpha release. Public APIs may still evolve before the stable 0.2.0 release.