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 ## Context
Syndarix agents require access to large language models (LLMs) from multiple providers: Syndarix agents require access to large language models (LLMs) from multiple providers:
- **Anthropic** (Claude) - Primary provider - **Anthropic** (Claude Opus 4.5) - Primary provider, highest reasoning capability
- **OpenAI** (GPT-4) - Fallback provider - **Google** (Gemini 3 Pro/Flash) - Strong multimodal, fast inference
- **Local models** (Ollama/Llama) - Cost optimization, privacy - **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: We need a unified abstraction layer that provides:
- Consistent API across providers - Consistent API across providers
@@ -79,25 +81,33 @@ LiteLLM provides the reliability, monitoring, and multi-provider support needed
### Model Groups ### Model Groups
| Group Name | Use Case | Primary Model | Fallback | | Group Name | Use Case | Primary Model | Fallback Chain |
|------------|----------|---------------|----------| |------------|----------|---------------|----------------|
| `high-reasoning` | Complex analysis, architecture | Claude 3.5 Sonnet | GPT-4 Turbo | | `high-reasoning` | Complex analysis, architecture | Claude Opus 4.5 | GPT 5.1 Codex max → Gemini 3 Pro |
| `fast-response` | Quick tasks, simple queries | Claude 3 Haiku | GPT-4o Mini | | `code-generation` | Code writing, refactoring | GPT 5.1 Codex max | Claude Opus 4.5 → DeepSeek V3.2 |
| `cost-optimized` | High-volume, non-critical | Local Llama 3 | Claude 3 Haiku | | `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) ▼ (on failure)
GPT-4 Turbo (OpenAI) DeepSeek V3.2 (Self-hosted)
│ │
▼ (on failure) ▼ (all failed)
Llama 3 (Ollama/Local) Error with exponential backoff retry
│
▼ (on failure)
Error with retry
``` ```
### LLM Gateway Service ### LLM Gateway Service
@@ -131,24 +141,26 @@ class LLMGateway:
### Cost Tracking ### Cost Tracking
| Model | Input (per 1M tokens) | Output (per 1M tokens) | | Model | Input (per 1M tokens) | Output (per 1M tokens) | Notes |
|-------|----------------------|------------------------| |-------|----------------------|------------------------|-------|
| Claude 3.5 Sonnet | $3.00 | $15.00 | | Claude Opus 4.5 | $15.00 | $75.00 | Highest reasoning capability |
| Claude 3 Haiku | $0.25 | $1.25 | | GPT 5.1 Codex max | $12.00 | $60.00 | Code generation specialist |
| GPT-4 Turbo | $10.00 | $30.00 | | Gemini 3 Pro | $3.50 | $10.50 | Strong multimodal |
| GPT-4o Mini | $0.15 | $0.60 | | Gemini 3 Flash | $0.35 | $1.05 | Fast inference |
| Ollama (local) | $0.00 | $0.00 | | 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 Mapping
| Agent Type | Model Preference | Rationale | | Agent Type | Model Preference | Rationale |
|------------|------------------|-----------| |------------|------------------|-----------|
| Product Owner | high-reasoning | Complex requirements analysis | | Product Owner | high-reasoning | Complex requirements analysis needs Claude Opus 4.5 |
| Software Architect | high-reasoning | Architecture decisions | | Software Architect | high-reasoning | Architecture decisions need top-tier reasoning |
| Software Engineer | high-reasoning | Code generation | | Software Engineer | code-generation | GPT 5.1 Codex max optimized for code |
| QA Engineer | fast-response | Test case generation | | QA Engineer | code-generation | Test code generation |
| DevOps Engineer | fast-response | Config generation | | DevOps Engineer | fast-response | Config generation (Gemini 3 Flash) |
| Project Manager | fast-response | Status updates | | Project Manager | fast-response | Status updates, quick responses |
| Business Analyst | high-reasoning | Document analysis needs strong reasoning |
### Caching Strategy ### 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 # 4. LiteLLM handles LLM calls with failover
async def think_node(state: AgentState) -> AgentState: async def think_node(state: AgentState) -> AgentState:
response = await litellm.acompletion( 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"], 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"]}, metadata={"agent_id": state["agent_id"]},
) )
return {"messages": [response.choices[0].message]} 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 ### Cost Model Prices
| Model | Input ($/1M) | Output ($/1M) | | Model | Input ($/1M) | Output ($/1M) | Notes |
|-------|-------------|---------------| |-------|-------------|---------------|-------|
| Claude 3.5 Sonnet | $3.00 | $15.00 | | Claude Opus 4.5 | $15.00 | $75.00 | Highest reasoning capability |
| Claude 3 Haiku | $0.25 | $1.25 | | GPT 5.1 Codex max | $12.00 | $60.00 | Code generation specialist |
| GPT-4 Turbo | $10.00 | $30.00 | | Gemini 3 Pro | $3.50 | $10.50 | Strong multimodal |
| GPT-4o Mini | $0.15 | $0.60 | | Gemini 3 Flash | $0.35 | $1.05 | Fast inference |
| Ollama (local) | $0.00 | $0.00 | | 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 ### Cost Optimization Strategies
| Strategy | Savings | Implementation | | Strategy | Savings | Implementation |
|----------|---------|----------------| |----------|---------|----------------|
| Semantic caching | 15-30% | Redis cache for repeated queries | | 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 | | 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 ## Consequences

52
docs/architecture/ARCHITECTURE.md
View File

@@ -42,10 +42,11 @@ Syndarix is an autonomous AI-powered software consulting platform that orchestra
│ │ │ │ │ │ │ │
│ │ ┌─────────────────────────────────────────────────────────────────────┐ │ │ │ │ ┌─────────────────────────────────────────────────────────────────────┐ │ │
│ │ │ ORCHESTRATION LAYER │ │ │ │ │ │ ORCHESTRATION LAYER │ │ │
│ │ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ │ │ │ │ │ │ ┌─────────────┐ ┌─────────────┐ ┌─────────────┐ ┌───────────┐ │ │ │
│ │ │ │ Agent │ │ Workflow │ │ Approval │ │ │ │ │ │ │ │ Agent │ │ Workflow │ │ Approval │ │ LangGraph │ │ │ │
│ │ │ │ Orchestrator│ │ Engine │ │ Service │ │ │ │ │ │ │ │ Orchestrator│ │ Engine │ │ Service │ │ Runtime │ │ │ │
│ │ │ └─────────────┘ └─────────────┘ └─────────────┘ │ │ │ │ │ │ │(Type-Inst.) │ │(transitions)│ │ │ │ │ │ │ │
│ │ │ └─────────────┘ └─────────────┘ └─────────────┘ └───────────┘ │ │ │
│ │ └─────────────────────────────────────────────────────────────────────┘ │ │ │ │ └─────────────────────────────────────────────────────────────────────┘ │ │
│ │ │ │ │ │ │ │
│ │ ┌─────────────────────────────────────────────────────────────────────┐ │ │ │ │ ┌─────────────────────────────────────────────────────────────────────┐ │ │
@@ -126,21 +127,26 @@ Agent Type (Template) Agent Instance (Runtime)
**Failover Chain:** **Failover Chain:**
``` ```
Claude 3.5 Sonnet (Primary) Claude Opus 4.5 (Primary)
│
▼ (on failure/rate limit)
GPT 5.1 Codex max (Code specialist)
│
▼ (on failure/rate limit)
Gemini 3 Pro (Multimodal)
│ │
▼ (on failure) ▼ (on failure)
GPT-4 Turbo (Fallback) Qwen3-235B / DeepSeek V3.2 (Self-hosted)
│
▼ (on failure)
Ollama/Llama 3 (Local)
``` ```
**Model Groups:** **Model Groups:**
| Group | Use Case | Primary Model | | Group | Use Case | Primary Model | Fallback |
|-------|----------|---------------| |-------|----------|---------------|----------|
| high-reasoning | Architecture, complex analysis | Claude 3.5 Sonnet | | high-reasoning | Architecture, complex analysis | Claude Opus 4.5 | GPT 5.1 Codex max |
| fast-response | Quick tasks, status updates | Claude 3 Haiku | | code-generation | Code writing, refactoring | GPT 5.1 Codex max | Claude Opus 4.5 |
| cost-optimized | High-volume, non-critical | Local Llama 3 | | fast-response | Quick tasks, status updates | Gemini 3 Flash | Qwen3-235B |
| cost-optimized | High-volume, non-critical | Qwen3-235B | DeepSeek V3.2 |
| self-hosted | Privacy-sensitive, air-gapped | DeepSeek V3.2 | Qwen3-235B |
### 3. Knowledge Base (RAG) ### 3. Knowledge Base (RAG)
@@ -245,13 +251,17 @@ LLM Request → LiteLLM Callback → Redis INCR → Budget Check
**All components are fully self-hostable with no mandatory subscriptions:** **All components are fully self-hostable with no mandatory subscriptions:**
| Component | Self-Hosted | Managed Alternative (Optional) | | Component | License | Self-Hosted | Managed Alternative (Optional) |
|-----------|-------------|--------------------------------| |-----------|---------|-------------|--------------------------------|
| PostgreSQL | Yes | RDS, Neon, Supabase | | PostgreSQL | PostgreSQL | Yes | RDS, Neon, Supabase |
| Redis | Yes | Redis Cloud | | Redis | BSD-3 | Yes | Redis Cloud |
| LiteLLM | Yes | LiteLLM Enterprise | | LiteLLM | MIT | Yes | LiteLLM Enterprise |
| Celery | Yes | - | | Celery | BSD-3 | Yes | - |
| FastMCP | Yes | - | | FastMCP | MIT | Yes | - |
| LangGraph | MIT | Yes | LangSmith (observability only) |
| transitions | MIT | Yes | - |
| DeepSeek V3.2 | MIT | Yes | API available |
| Qwen3-235B | Apache 2.0 | Yes | Alibaba Cloud |
--- ---

19
docs/architecture/IMPLEMENTATION_ROADMAP.md
View File

@@ -295,10 +295,13 @@ This roadmap outlines the phased implementation approach for Syndarix, prioritiz
| Metric | Target | Measurement | | Metric | Target | Measurement |
|--------|--------|-------------| |--------|--------|-------------|
| Agent task success rate | >90% | Completed tasks / total tasks | | Agent task success rate | >90% | Completed tasks / total tasks |
| Response time (P95) | <2s | API latency | | API response time (P95) | <200ms | Pure API latency (per NFR-101) |
| Cost per project | <$50/sprint | LLM + compute costs | | Agent response time | <10s simple, <60s code | End-to-end including LLM (per NFR-103) |
| Cost per project | <$100/sprint | LLM + compute costs (with Opus 4.5 pricing) |
| Time to first commit | <1 hour | From requirements to PR | | Time to first commit | <1 hour | From requirements to PR |
| Client satisfaction | >4/5 | Post-sprint survey | | Client satisfaction | >4/5 | Post-sprint survey |
| Concurrent projects | 10+ | Active projects in parallel |
| Concurrent agents | 50+ | Agent instances running |
--- ---
@@ -328,15 +331,17 @@ Foundation Core Platform MCP Integration Agent Orch Workflows Advan
### Infrastructure ### Infrastructure
- PostgreSQL (managed or self-hosted) - PostgreSQL (managed or self-hosted)
- Redis (managed or self-hosted) - Redis (managed or self-hosted)
- Celery workers (2-4 instances) - Celery workers (4-8 instances across 4 queues: agent, git, sync, cicd)
- MCP servers (7 containers) - MCP servers (7 containers, deployed in Phase 2 + Phase 5)
- API server (2+ instances) - API server (2+ instances)
- Frontend (static hosting or SSR) - Frontend (static hosting or SSR)
### External Services ### External Services
- Anthropic API (primary LLM) - Anthropic API (Claude Opus 4.5 - primary reasoning)
- OpenAI API (fallback) - OpenAI API (GPT 5.1 Codex max - code generation)
- Ollama (local models, optional) - Google API (Gemini 3 Pro/Flash - multimodal, fast)
- Alibaba API (Qwen3-235B - cost-effective, or self-host)
- DeepSeek V3.2 (self-hosted, open weights)
- Gitea/GitHub/GitLab (issue tracking) - Gitea/GitHub/GitLab (issue tracking)
--- ---