AgentForge

Appearance

Changelog ​

All notable changes to AgentForge.

[1.0.0] - 2025-01-15 ​

🎉 Production Release ​

AgentForge v1.0.0 is here! A production-ready TypeScript framework for building AI agents with enterprise-grade reliability.

Core Features ​

Agent System ​

  • Agent class - Orchestrates conversations with LLM providers
  • Tool system - Type-safe tools with Zod validation
  • Streaming - Real-time token streaming with async iterators
  • Memory management - Configurable conversation history strategies

Providers ​

  • OpenAI - Full support for GPT-4, GPT-4 Turbo, and GPT-3.5 models
  • Anthropic - Full support for Claude 3 and Claude 3.5 models
  • Custom providers - Extensible base class for any LLM
  • Provider Factory - Dynamic provider creation and configuration
  • Multi-Provider - Load balancing across multiple providers
  • Failover Provider - Automatic fallback on provider failures

Middleware Pipeline ​

  • createMiddleware - Build custom middleware
  • loggingMiddleware - Built-in request/response logging
  • createRateLimitMiddleware - Configurable rate limiting
  • createCacheMiddleware - Response caching with TTL
  • createRetryMiddleware - Automatic retries with backoff

React Integration ​

  • useAgent - Full-featured agent hook
  • useChat - Simplified chat interface hook
  • useStreamingAgent - Real-time streaming support
  • AgentProvider - Context provider for app-wide configuration

Enterprise Features ​

Resilience Patterns ​

  • CircuitBreaker - Prevent cascade failures
  • Bulkhead - Isolate failures between components
  • RequestDeduplicator - Eliminate duplicate concurrent requests
  • retryWithBackoff - Exponential backoff retry strategy
  • withTimeout - Request timeout handling
  • withFallback - Graceful fallback responses
  • HealthChecker - Monitor provider health status

Request Interceptors ​

  • InterceptorChain - Composable request/response pipeline
  • createLoggingInterceptor - Request/response logging
  • createMetricsInterceptor - Performance metrics collection
  • createContentFilterInterceptor - Sensitive content filtering
  • createInjectionDetectionInterceptor - Prompt injection protection
  • createTransformInterceptor - Request/response transformation
  • createRetryErrorInterceptor - Error-based retry logic
  • createCircuitBreakerErrorInterceptor - Circuit breaker integration

Graceful Degradation ​

  • DegradationManager - Manage service degradation levels
  • Feature Flags - Toggle capabilities dynamically
  • createDegradedTool - Fallback tool implementations
  • simplifyResponse - Reduce response complexity under load
  • Health-Aware Execution - Adapt behavior to system health

Observability ​

  • TelemetryCollector - Comprehensive metrics collection
  • Distributed Tracing - Span-based request tracing
  • OTLP Export - OpenTelemetry Protocol support
  • Structured Logging - JSON logging with log levels
  • Custom Transports - Console, file, and batching transports

Performance ​

  • RequestBatcher - Batch multiple requests for efficiency
  • RequestDeduplicator - Coalesce identical concurrent requests
  • RateLimitedQueue - Queue-based rate limiting
  • Token Counting - Accurate token estimation per model
  • Context Window Management - Automatic message truncation

Persistence ​

  • ConversationManager - Save and restore conversations
  • MemoryStorageAdapter - In-memory storage for development
  • FileStorageAdapter - File-based persistent storage
  • Conversation Search - Full-text conversation search

Type Safety ​

Error Handling ​

  • Typed errors - AgentForgeError, ProviderError, ToolExecutionError, ValidationError
  • Error codes - Comprehensive error code enum
  • Result types - Functional error handling (Ok/Err pattern)
  • Error context - Rich error metadata

Branded Types ​

  • MessageId, ToolCallId, ResponseId - Type-safe identifiers
  • Temperature, TokenCount - Constrained numeric types
  • NonEmptyString, ToolName, ApiKey - Validated string types

Type Guards ​

  • isUserMessage, isAssistantMessage - Message type narrowing
  • hasToolCalls, isToolResult - Tool-related guards
  • assertDefined, assertNonEmptyString - Runtime assertions

Schema Validation ​

  • Zod schemas - CompletionResponse, StreamChunk, Message schemas
  • validateOrThrow - Strict validation with errors
  • sanitizeCompletionResponse - Clean and normalize responses
  • createToolValidator - Dynamic tool validation

Documentation ​

  • Complete API reference
  • Getting started guide
  • Core concepts explanation
  • Real-world examples (Customer Support, Data Analyst, Code Assistant)
  • Interactive playground

[1.1.0] - 2025-01-15 Completed ​

🚀 New Features Release ​

Major expansion of AgentForge with new providers, memory strategies, and UI components.

New Providers ​

Google Gemini ​

  • GeminiProvider - Full support for Gemini Pro, Gemini 1.5 Pro, and Gemini 1.5 Flash
  • Native function calling support
  • System instruction handling
  • Streaming with Server-Sent Events

Cohere ​

  • CohereProvider - Support for Command, Command-R, and Command-R+ models
  • Tool/function calling support
  • Preamble-based system messages
  • NDJSON streaming

Together AI ​

  • TogetherProvider - Access to 100+ open-source models
  • TOGETHER_MODELS - Pre-configured model constants for popular models
  • Support for Llama 3.1, Mixtral, Mistral, DeepSeek, Qwen, and more
  • Full OpenAI-compatible API

Memory Summarization Strategies ​

New intelligent memory management for long conversations:

  • createSlidingWindowSummarizer - Simple sliding window with optional summarization of dropped messages
  • createSemanticCompressionSummarizer - LLM-powered compression preserving key information
  • createHierarchicalSummarizer - Multi-level summaries (recent detail, mid-term context, long-term summary)
  • createImportanceBasedSummarizer - Retain messages based on importance scoring
  • createMemorySummarizer - Factory function for easy strategy selection

Enhanced React Components ​

Production-ready UI components for building chat interfaces:

  • ChatWindow - Complete chat interface with input, auto-scroll, streaming support
  • MessageList - Scrollable message display with custom renderers
  • MessageBubble - Individual message rendering with code block support
  • ToolStatus - Tool execution visualization with progress tracking
  • TypingIndicator - Animated loading/streaming indicator
  • ChatErrorBoundary - Error boundary for graceful error handling
  • Avatar - Configurable avatar component for messages

Roadmap ​

[1.2.0] - Planned ​

  • [ ] WebSocket streaming transport
  • [ ] Conversation branching and forking
  • [ ] Tool dependency management
  • [ ] Multi-agent orchestration patterns
  • [ ] Agent-to-agent communication
  • [ ] Workflow engine for complex task pipelines

[1.3.0] - Planned ​

  • [ ] Built-in observability dashboard
  • [ ] Edge runtime support (Cloudflare Workers, Vercel Edge)
  • [ ] Vector store integrations (Pinecone, Weaviate, Chroma)
  • [ ] Fine-tuning integration utilities
  • [ ] Prompt versioning and A/B testing

[1.4.0] - Planned ​

  • [ ] Cost tracking and budgeting
  • [ ] Compliance logging (audit trails)
  • [ ] Multi-modal support (images, audio)
  • [ ] Advanced caching strategies
  • [ ] Conversation analytics

Contributing ​

See CONTRIBUTING.md for guidelines.

License ​

MIT © Michael Palmer

Released under the MIT License.

Copyright © 2026 Michael Palmer