@strav/brain 0.4.30 → 1.0.0-alpha.10

package/CHANGELOG.md

@@ -1,44 +0,0 @@
-# Changelog
-
-## 0.2.12
-
-### Added
-
-- **GoogleProvider** — Support for Google's Gemini models
-  - Native Gemini API integration using `generativelanguage.googleapis.com`
-  - Support for completion, streaming, function calling, and embeddings
-  - Models: `gemini-2.0-flash`, `gemini-2.5-flash`, `gemini-3-pro-preview`
-  - Authentication via `x-goog-api-key` header
-  - Zero new dependencies — uses raw `fetch()` following existing patterns
-  - Comprehensive test suite with 29 tests covering all functionality
-
-## 0.6.0
-
-### Added
-
-- **Memory management** — three-tier conversation memory system for long-running threads
-  - `thread.memory()` enables opt-in context window management
-  - **Working memory** — recent messages within token budget
-  - **Episodic memory** — LLM-generated summaries of compacted older messages
-  - **Semantic memory** — structured facts extracted from conversation, injected into system prompt
-- `TokenCounter` — approximate token estimation per provider (~4 chars/token)
-- `ContextBudget` — budget allocation across system prompt, summaries, facts, and working messages
-- `MemoryManager` — orchestrates compaction and fact extraction
-- `SemanticMemory` — in-memory fact store with `<known_facts>` prompt injection
-- `SummarizeStrategy` — LLM-powered compaction with optional fact extraction
-- `SlidingWindowStrategy` — drop oldest messages without summarization
-- `InMemoryThreadStore` — default `ThreadStore` implementation for dev/testing
-- `ThreadStore` interface — pluggable persistence (implement for database-backed storage)
-- `BrainManager.useThreadStore()` — register a thread store for persistence
-- `BrainManager.memoryConfig` / `BrainManager.threadStore` — accessors for memory configuration
-- `thread.id()` — set thread identifier for persistence
-- `thread.persist()` — enable auto-save to ThreadStore after each `send()`
-- `thread.facts` / `thread.episodicSummary` — access memory state
-- `thread.serializeMemory()` / `thread.restoreMemory()` — extended serialization with memory state
-- `BrainConfig.memory` — optional `MemoryConfig` field for global memory settings
-
-## 0.1.1
-
-### Changed
-
-- Applied consistent code formatting across all source files

package/README.md

@@ -1,121 +0,0 @@
-# @strav/brain
-
-AI module for the [Strav](https://www.npmjs.com/package/@strav/core) framework. Provides a unified interface for AI providers with support for agents, threads, tool use, and multi-step workflows.
-
-## Install
-
-```bash
-bun add @strav/brain
-```
-
-Requires `@strav/core` as a peer dependency.
-
-## Providers
-
-- **Anthropic** (Claude)
-- **OpenAI** (GPT, also works with DeepSeek via custom `baseUrl`)
-- **Google** (Gemini)
-
-## Usage
-
-```ts
-import { brain } from '@strav/brain'
-
-// One-shot chat
-const response = await brain.chat('Explain quantum computing')
-
-// Streaming
-for await (const chunk of brain.stream('Write a poem')) {
-  process.stdout.write(chunk.text)
-}
-
-// Structured output with Zod
-import { z } from 'zod'
-const result = await brain.generate('List 3 colors', {
-  schema: z.object({ colors: z.array(z.string()) }),
-})
-
-// Embeddings
-const vectors = await brain.embed('Hello world')
-```
-
-## Tools
-
-Define tools that AI agents can use:
-
-```ts
-import { defineTool } from '@strav/brain'
-import { z } from 'zod'
-
-const searchTool = defineTool({
-  name: 'search',
-  description: 'Search the database',
-  parameters: z.object({ query: z.string() }),
-  execute: async ({ query }, context) => {
-    const userId = context?.userId
-    return await db.search(query, { userId })
-  },
-})
-```
-
-The `execute` function receives two parameters:
-- `args` - The parsed and validated tool arguments
-- `context` - Optional context object passed from the agent runner
-
-## Agents
-
-```ts
-import { Agent, defineTool } from '@strav/brain'
-
-class ResearchAgent extends Agent {
-  provider = 'anthropic'
-  model = 'claude-sonnet-4-20250514'
-  instructions = 'You are a research assistant.'
-  tools = [searchTool, summarizeTool]
-}
-
-// Google Gemini agent
-class GeminiResearchAgent extends Agent {
-  provider = 'google'
-  model = 'gemini-2.0-flash'
-  instructions = 'You are a research assistant powered by Gemini.'
-  tools = [searchTool, summarizeTool]
-}
-
-// Run agent with context
-const runner = brain.agent(ResearchAgent)
-runner.context({ userId: '123' }) // Pass context to tools
-const result = await runner.input('Find info on Bun').run()
-```
-
-## Threads
-
-Multi-turn conversations with serialization support:
-
-```ts
-const thread = brain.thread({ provider: 'anthropic', model: 'claude-sonnet-4-20250514' })
-await thread.send('Hello')
-await thread.send('Tell me more')
-const saved = thread.serialize() // persist and restore later
-
-// Google Gemini example
-const geminiThread = brain.thread({ provider: 'google', model: 'gemini-2.0-flash' })
-await geminiThread.send('Explain quantum computing')
-```
-
-## Workflows
-
-Orchestrate multi-agent pipelines:
-
-```ts
-const workflow = brain.workflow()
-  .step('research', ResearchAgent)
-  .step('summarize', SummaryAgent)
-  .parallel('review', [FactCheckAgent, StyleAgent])
-
-const result = await workflow.run('Analyze this topic')
-```
-
-## License
-
-MIT