cardosofelipe/syndarix
1
0
Fork 0
forked from cardosofelipe/fast-next-template
Code Issues 18 Pull Requests Wiki Activity
503 Commits 3 Branches 0 Tags
4ad3d20cf2f23d9a0d0880af332ed5f573f3c26e
Commit Graph

10 Commits

Author SHA1 Message Date
Felipe Cardoso
30e5c68304 feat(memory): integrate memory system with context engine (#97) 
## Changes

### New Context Type
- Add MEMORY to ContextType enum for agent memory context
- Create MemoryContext class with subtypes (working, episodic, semantic, procedural)
- Factory methods: from_working_memory, from_episodic_memory, from_semantic_memory, from_procedural_memory

### Memory Context Source
- MemoryContextSource service fetches relevant memories for context assembly
- Configurable fetch limits per memory type
- Parallel fetching from all memory types

### Agent Lifecycle Hooks
- AgentLifecycleManager handles spawn, pause, resume, terminate events
- spawn: Initialize working memory with optional initial state
- pause: Create checkpoint of working memory
- resume: Restore from checkpoint
- terminate: Consolidate working memory to episodic memory
- LifecycleHooks for custom extension points

### Context Engine Integration
- Add memory_query parameter to assemble_context()
- Add session_id and agent_type_id for memory scoping
- Memory budget allocation (15% by default)
- set_memory_source() for runtime configuration

### Tests
- 48 new tests for MemoryContext, MemoryContextSource, and lifecycle hooks
- All 108 memory-related tests passing
- mypy and ruff checks passing

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
 2026-01-05 03:49:22 +01:00
Felipe Cardoso
2bea057fb1 chore(context): refactor for consistency, optimize formatting, and simplify logic 
- Cleaned up unnecessary comments in `__all__` definitions for better readability.
- Adjusted indentation and formatting across modules for improved clarity (e.g., long lines, logical grouping).
- Simplified conditional expressions and inline comments for context scoring and ranking.
- Replaced some hard-coded values with type-safe annotations (e.g., `ClassVar`).
- Removed unused imports and ensured consistent usage across test files.
- Updated `test_score_not_cached_on_context` to clarify caching behavior.
- Improved truncation strategy logic and marker handling.
 2026-01-04 15:23:14 +01:00
Felipe Cardoso
6c7b72f130 chore(context): apply linter fixes and sort imports (#86) 
Phase 8 of Context Management Engine - Final Cleanup:

- Sort __all__ exports alphabetically
- Sort imports per isort conventions
- Fix minor linting issues

Final test results:
- 311 context management tests passing
- 2507 total backend tests passing
- 85% code coverage

Context Management Engine is complete with all 8 phases:
1. Foundation: Types, Config, Exceptions
2. Token Budget Management
3. Context Scoring & Ranking
4. Context Assembly Pipeline
5. Model Adapters (Claude, OpenAI)
6. Caching Layer (Redis + in-memory)
7. Main Engine & Integration
8. Testing & Documentation

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
 2026-01-04 02:46:56 +01:00
Felipe Cardoso
027ebfc332 feat(context): implement main ContextEngine with full integration (#85) 
Phase 7 of Context Management Engine - Main Engine:

- Add ContextEngine as main orchestration class
- Integrate all components: calculator, scorer, ranker, compressor, cache
- Add high-level assemble_context() API with:
  - System prompt support
  - Task description support
  - Knowledge Base integration via MCP
  - Conversation history conversion
  - Tool results conversion
  - Custom contexts support
- Add helper methods:
  - get_budget_for_model()
  - count_tokens() with caching
  - invalidate_cache()
  - get_stats()
- Add create_context_engine() factory function

Tests: 26 new tests, 311 total context tests passing

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
 2026-01-04 02:44:40 +01:00
Felipe Cardoso
c2466ab401 feat(context): implement Redis-based caching layer (#84) 
Phase 6 of Context Management Engine - Caching Layer:

- Add ContextCache with Redis integration
- Support fingerprint-based assembled context caching
- Support token count caching (model-specific)
- Support score caching (scorer + context + query)
- Add in-memory fallback with LRU eviction
- Add cache invalidation with pattern matching
- Add cache statistics reporting

Key features:
- Hierarchical cache key structure (ctx:type:hash)
- Automatic TTL expiration
- Memory cache for fast repeated access
- Graceful degradation when Redis unavailable

Tests: 29 new tests, 285 total context tests passing

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
 2026-01-04 02:41:21 +01:00
Felipe Cardoso
7828d35e06 feat(context): implement model adapters for Claude and OpenAI (#83) 
Phase 5 of Context Management Engine - Model Adapters:

- Add ModelAdapter abstract base class with model matching
- Add DefaultAdapter for unknown models (plain text)
- Add ClaudeAdapter with XML-based formatting:
  - <system_instructions> for system context
  - <reference_documents>/<document> for knowledge
  - <conversation_history>/<message> for chat
  - <tool_results>/<tool_result> for tool outputs
  - XML escaping for special characters
- Add OpenAIAdapter with markdown formatting:
  - ## headers for sections
  - ### Source headers for documents
  - **ROLE** bold labels for conversation
  - Code blocks for tool outputs
- Add get_adapter() factory function for model selection

Tests: 33 new tests, 256 total context tests passing

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
 2026-01-04 02:36:32 +01:00
Felipe Cardoso
6b07e62f00 feat(context): implement assembly pipeline and compression (#82) 
Phase 4 of Context Management Engine - Assembly Pipeline:

- Add TruncationStrategy with end/middle/sentence-aware truncation
- Add TruncationResult dataclass for tracking compression metrics
- Add ContextCompressor for type-specific compression
- Add ContextPipeline orchestrating full assembly workflow:
  - Token counting for all contexts
  - Scoring and ranking via ContextRanker
  - Optional compression when budget threshold exceeded
  - Model-specific formatting (XML for Claude, markdown for OpenAI)
- Add PipelineMetrics for performance tracking
- Update AssembledContext with new fields (model, contexts, metadata)
- Add backward compatibility aliases for renamed fields

Tests: 34 new tests, 223 total context tests passing

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
 2026-01-04 02:32:25 +01:00
Felipe Cardoso
0d2005ddcb feat(context): implement context scoring and ranking (Phase 3) 
Add comprehensive scoring system with three strategies:
- RelevanceScorer: Semantic similarity with keyword fallback
- RecencyScorer: Exponential decay with type-specific half-lives
- PriorityScorer: Priority-based scoring with type bonuses

Implement CompositeScorer combining all strategies with configurable
weights (default: 50% relevance, 30% recency, 20% priority).

Add ContextRanker for budget-aware context selection with:
- Greedy selection algorithm respecting token budgets
- CRITICAL priority contexts always included
- Diversity reranking to prevent source dominance
- Comprehensive selection statistics

68 tests covering all scoring and ranking functionality.

Part of #61 - Context Management Engine

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
 2026-01-04 02:24:06 +01:00
Felipe Cardoso
dfa75e682e feat(context): implement token budget management (Phase 2) 
Add TokenCalculator with LLM Gateway integration for accurate token
counting with in-memory caching and fallback character-based estimation.
Implement TokenBudget for tracking allocations per context type with
budget enforcement, and BudgetAllocator for creating budgets based on
model context window sizes.

- TokenCalculator: MCP integration, caching, model-specific ratios
- TokenBudget: allocation tracking, can_fit/allocate/deallocate/reset
- BudgetAllocator: model context sizes, budget creation and adjustment
- 35 comprehensive tests covering all budget functionality

Part of #61 - Context Management Engine

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
 2026-01-04 02:13:23 +01:00
Felipe Cardoso
22ecb5e989 feat(context): Phase 1 - Foundation types, config and exceptions (#79) 
Implements the foundation for Context Management Engine:

Types (backend/app/services/context/types/):
- BaseContext: Abstract base with ID, content, priority, scoring
- SystemContext: System prompts, personas, instructions
- KnowledgeContext: RAG results from Knowledge Base MCP
- ConversationContext: Chat history with role support
- TaskContext: Task/issue context with acceptance criteria
- ToolContext: Tool definitions and execution results
- AssembledContext: Final assembled context result

Configuration (config.py):
- Token budget allocation (system 5%, task 10%, knowledge 40%, etc.)
- Scoring weights (relevance 50%, recency 30%, priority 20%)
- Cache settings (TTL, prefix)
- Performance settings (max assembly time, parallel scoring)
- Environment variable overrides with CTX_ prefix

Exceptions (exceptions.py):
- ContextError: Base exception
- BudgetExceededError: Token budget violations
- TokenCountError: Token counting failures
- CompressionError: Compression failures
- AssemblyTimeoutError: Assembly timeout
- ScoringError, FormattingError, CacheError
- ContextNotFoundError, InvalidContextError

All 86 tests pass.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
 2026-01-04 02:07:39 +01:00