cardosofelipe/syndarix
1
0
Fork 0
forked from cardosofelipe/fast-next-template
Code Issues 18 Pull Requests Wiki Activity

feat: Update to production model stack and fix remaining inconsistencies

Browse Source 
## Model Stack Updates (User's Actual Models)

Updated all documentation to reflect production models:
- Claude Opus 4.5 (primary reasoning)
- GPT 5.1 Codex max (code generation specialist)
- Gemini 3 Pro/Flash (multimodal, fast inference)
- Qwen3-235B (cost-effective, self-hostable)
- DeepSeek V3.2 (self-hosted, open weights)

### Files Updated:
- ADR-004: Full model groups, failover chains, cost tables
- ADR-007: Code example with correct model identifiers
- ADR-012: Cost tracking with new model prices
- ARCHITECTURE.md: Model groups, failover diagram
- IMPLEMENTATION_ROADMAP.md: External services list

## Architecture Diagram Updates

- Added LangGraph Runtime to orchestration layer
- Added technology labels (Type-Instance, transitions)

## Self-Hostability Table Expanded

Added entries for:
- LangGraph (MIT)
- transitions (MIT)
- DeepSeek V3.2 (MIT)
- Qwen3-235B (Apache 2.0)

## Metric Alignments

- Response time: Split into API (<200ms) and Agent (<10s/<60s)
- Cost per project: Adjusted to $100/sprint for Opus 4.5 pricing
- Added concurrent projects (10+) and agents (50+) metrics

## Infrastructure Updates

- Celery workers: 4-8 instances (was 2-4) across 4 queues
- MCP servers: Clarified Phase 2 + Phase 5 deployment
- Sync interval: Clarified 60s fallback + 15min reconciliation

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

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Felipe Cardoso
2025-12-29 23:35:51 +01:00
parent f138417486
commit 88cf4e0abc
6 changed files with 98 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

70
docs/adrs/ADR-004-llm-provider-abstraction.md
View File

@@ -10,9 +10,11 @@
## Context
Syndarix agents require access to large language models (LLMs) from multiple providers:
- **Anthropic** (Claude) - Primary provider
- **OpenAI** (GPT-4) - Fallback provider
- **Local models** (Ollama/Llama) - Cost optimization, privacy
- **Anthropic** (Claude Opus 4.5) - Primary provider, highest reasoning capability
- **Google** (Gemini 3 Pro/Flash) - Strong multimodal, fast inference
- **OpenAI** (GPT 5.1 Codex max) - Code generation specialist
- **Alibaba** (Qwen3-235B) - Cost-effective alternative
- **DeepSeek** (V3.2) - Open-weights, self-hostable option
We need a unified abstraction layer that provides:
- Consistent API across providers
@@ -79,25 +81,33 @@ LiteLLM provides the reliability, monitoring, and multi-provider support needed
### Model Groups
| Group Name | Use Case | Primary Model | Fallback |
|------------|----------|---------------|----------|
| `high-reasoning` | Complex analysis, architecture | Claude 3.5 Sonnet | GPT-4 Turbo |
| `fast-response` | Quick tasks, simple queries | Claude 3 Haiku | GPT-4o Mini |
| `cost-optimized` | High-volume, non-critical | Local Llama 3 | Claude 3 Haiku |
| Group Name | Use Case | Primary Model | Fallback Chain |
|------------|----------|---------------|----------------|
| `high-reasoning` | Complex analysis, architecture | Claude Opus 4.5 | GPT 5.1 Codex max → Gemini 3 Pro |
| `code-generation` | Code writing, refactoring | GPT 5.1 Codex max | Claude Opus 4.5 → DeepSeek V3.2 |
| `fast-response` | Quick tasks, simple queries | Gemini 3 Flash | Qwen3-235B → DeepSeek V3.2 |
| `cost-optimized` | High-volume, non-critical | Qwen3-235B | DeepSeek V3.2 (self-hosted) |
| `self-hosted` | Privacy-sensitive, air-gapped | DeepSeek V3.2 | Qwen3-235B |
### Failover Chain
### Failover Chain (Primary)
```
Claude 3.5 Sonnet (Anthropic)
Claude Opus 4.5 (Anthropic)
│
▼ (on failure/rate limit)
GPT 5.1 Codex max (OpenAI)
│
▼ (on failure/rate limit)
Gemini 3 Pro (Google)
│
▼ (on failure/rate limit)
Qwen3-235B (Alibaba/Self-hosted)
│
▼ (on failure)
GPT-4 Turbo (OpenAI)
DeepSeek V3.2 (Self-hosted)
│
▼ (on failure)
Llama 3 (Ollama/Local)
│
▼ (on failure)
Error with retry
▼ (all failed)
Error with exponential backoff retry
```
### LLM Gateway Service
@@ -131,24 +141,26 @@ class LLMGateway:
### Cost Tracking
| Model | Input (per 1M tokens) | Output (per 1M tokens) |
|-------|----------------------|------------------------|
| Claude 3.5 Sonnet | $3.00 | $15.00 |
| Claude 3 Haiku | $0.25 | $1.25 |
| GPT-4 Turbo | $10.00 | $30.00 |
| GPT-4o Mini | $0.15 | $0.60 |
| Ollama (local) | $0.00 | $0.00 |
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Notes |
|-------|----------------------|------------------------|-------|
| Claude Opus 4.5 | $15.00 | $75.00 | Highest reasoning capability |
| GPT 5.1 Codex max | $12.00 | $60.00 | Code generation specialist |
| Gemini 3 Pro | $3.50 | $10.50 | Strong multimodal |
| Gemini 3 Flash | $0.35 | $1.05 | Fast inference |
| Qwen3-235B | $2.00 | $6.00 | Cost-effective (or self-host: $0) |
| DeepSeek V3.2 | $0.00 | $0.00 | Self-hosted, open weights |
### Agent Type Mapping
| Agent Type | Model Preference | Rationale |
|------------|------------------|-----------|
| Product Owner | high-reasoning | Complex requirements analysis |
| Software Architect | high-reasoning | Architecture decisions |
| Software Engineer | high-reasoning | Code generation |
| QA Engineer | fast-response | Test case generation |
| DevOps Engineer | fast-response | Config generation |
| Project Manager | fast-response | Status updates |
| Product Owner | high-reasoning | Complex requirements analysis needs Claude Opus 4.5 |
| Software Architect | high-reasoning | Architecture decisions need top-tier reasoning |
| Software Engineer | code-generation | GPT 5.1 Codex max optimized for code |
| QA Engineer | code-generation | Test code generation |
| DevOps Engineer | fast-response | Config generation (Gemini 3 Flash) |
| Project Manager | fast-response | Status updates, quick responses |
| Business Analyst | high-reasoning | Document analysis needs strong reasoning |
### Caching Strategy

4
docs/adrs/ADR-007-agentic-framework-selection.md
View File

@@ -297,9 +297,9 @@ def create_agent_graph() -> StateGraph:
# 4. LiteLLM handles LLM calls with failover
async def think_node(state: AgentState) -> AgentState:
response = await litellm.acompletion(
model="claude-3-5-sonnet-latest", # Claude 3.5 Sonnet (primary)
model="claude-opus-4-5", # Claude Opus 4.5 (primary)
messages=state["messages"],
fallbacks=["gpt-4-turbo", "ollama/llama3"],
fallbacks=["gpt-5.1-codex-max", "gemini-3-pro", "qwen3-235b", "deepseek-v3.2"],
metadata={"agent_id": state["agent_id"]},
)
return {"messages": [response.choices[0].message]}

2
docs/adrs/ADR-011-issue-synchronization.md
View File

@@ -61,7 +61,7 @@ External Trackers (Gitea/GitHub/GitLab)
└───────────────────┘
│
┌─────────┴─────────┐
│ Polling Worker │ (reconciliation every 15 min)
│ Polling Worker │ (fallback: 60s, full reconciliation: 15 min)
└───────────────────┘
│
┌─────────┴─────────┐

20
docs/adrs/ADR-012-cost-tracking.md
View File

@@ -154,22 +154,24 @@ GROUP BY project_id, DATE(timestamp);
### Cost Model Prices
| Model | Input ($/1M) | Output ($/1M) |
|-------|-------------|---------------|
| Claude 3.5 Sonnet | $3.00 | $15.00 |
| Claude 3 Haiku | $0.25 | $1.25 |
| GPT-4 Turbo | $10.00 | $30.00 |
| GPT-4o Mini | $0.15 | $0.60 |
| Ollama (local) | $0.00 | $0.00 |
| Model | Input ($/1M) | Output ($/1M) | Notes |
|-------|-------------|---------------|-------|
| Claude Opus 4.5 | $15.00 | $75.00 | Highest reasoning capability |
| GPT 5.1 Codex max | $12.00 | $60.00 | Code generation specialist |
| Gemini 3 Pro | $3.50 | $10.50 | Strong multimodal |
| Gemini 3 Flash | $0.35 | $1.05 | Fast inference |
| Qwen3-235B | $2.00 | $6.00 | Cost-effective (or $0 self-hosted) |
| DeepSeek V3.2 | $0.00 | $0.00 | Self-hosted, open weights |
### Cost Optimization Strategies
| Strategy | Savings | Implementation |
|----------|---------|----------------|
| Semantic caching | 15-30% | Redis cache for repeated queries |
| Model cascading | 60-80% | Start with Haiku, escalate to Sonnet |
| Model cascading | 60-80% | Start with Gemini Flash, escalate to Opus |
| Prompt compression | 10-20% | Remove redundant context |
| Local fallback | 100% for some | Ollama for simple tasks |
| Self-hosted fallback | 100% for some | DeepSeek V3.2/Qwen3 for non-critical tasks |
| Task-appropriate routing | 40-60% | Route code tasks to GPT 5.1 Codex, simple to Flash |
## Consequences