PostQode Provider & Model Selection Guide

Selecting the right AI model and provider is critical for balancing architectural reasoning depth, surgical execution accuracy, and token economy. PostQode gives you complete flexibility through either PostQode Managed Gateway or Bring Your Own Key (BYOK) across 37+ LLM providers.

---

📸 Model Configuration Interface

---

1. Managed Provider vs. Bring Your Own Key (BYOK)

⚡ PostQode Managed Provider

Zero-configuration AI gateway with automated multi-region failover and centralized token billing.

• Zero Key Management: No individual vendor API keys required
• Automated Failover: Intelligently routes traffic during outages
• Enterprise Governance: Model whitelisting, spend caps, and audit logs
• Zero Data Retention: Enterprise compliance — zero training on code

🔑 Bring Your Own Key (BYOK)

Connect directly to 37+ third-party cloud providers or your own private local models.

• Direct Vendor Billing: Use existing Anthropic, OpenAI, or AWS credits
• Self-Hosted Models: Run Ollama or LM Studio locally with 100% offline privacy
• Internal Enterprise Proxies: Connect custom LiteLLM or VNet endpoints
• Custom Temperature & Settings: Granular control per provider

---

2. Model Selection Matrix by Tier

🧠 Deep Reasoning & Planning Tier PLAN MODE

Engineered for deep repository analysis, multi-file architectural planning, and complex bug diagnosis.

• Claude 3.7 Sonnet (Thinking) — Frontier reasoning with thinking budgets
• DeepSeek-R1 — Open-weights multi-step mathematical & coding logic
• OpenAI o1 / o3-mini — High-depth chain-of-thought verification

⚡ Active Coding & Patching Tier AGENT MODE

Optimized for surgical unified diff generation, accurate tool calling, and high instruction following.

• Claude 3.5 Sonnet — Best balance of code quality, speed, and cost
• GPT-4o / GPT-5.5 — Fast, robust multimodal coding and refactoring
• Qwen 2.5 Coder 32B — Top open-source code generation model

🚀 Ultra-Fast Wafer-Scale Tier 1,000+ TOK/S

Near-instantaneous response times for real-time interactive pair programming and rapid iteration.

• Cerebras CS-3 (Llama 3.3 70B) — 1,000+ tokens/sec inference
• SambaNova Cloud — 16-bit full precision RDU inference
• Groq LPU — High-speed low-latency token streaming

🔒 Self-Hosted & Air-Gapped Tier 100% PRIVATE

Zero data egress for strict enterprise compliance or completely offline development.

• Ollama — DeepSeek-Coder, Qwen 2.5 Coder, Devstral on local GPU/CPU
• LM Studio — Local OpenAI-compatible server with custom GGUF models
• LiteLLM Proxy — Self-hosted private enterprise model gateway

---

3. Configuring Models per Execution Mode

PostQode allows you to pair different AI models for Plan Mode vs. Agent Mode to maximize efficiency:

                            User Task Request
                                    │
                                    ▼
       ┌────────────────────────────────────────────────────────┐
       │             Phase 1: Architecture & Planning           │
       │                   [ PLAN MODE ]                        │
       │       Model: Claude 3.7 Sonnet (Thinking) / R1         │
       │   • Deep repository scan via lsp_query & search_files  │
       │   • Identifies edge cases & drafts implementation plan │
       │   • 100% Read-Only (Zero risk of broken file edits)    │
       └────────────────────────────┬───────────────────────────┘
                                    │ Plan Approved by User
                                    ▼
       ┌────────────────────────────────────────────────────────┐
       │             Phase 2: Autonomous Execution              │
       │                   [ AGENT MODE ]                       │
       │           Model: Claude 3.5 Sonnet / GPT-4o            │
       │   • Applies surgical patches via apply_patch           │
       │   • Runs terminal test commands & linters              │
       │   • Completes task cleanly with commit diffs           │
       └────────────────────────────────────────────────────────┘

Settings UI

1. Open PostQode Settings ().
2. Under Model Configuration, select your Plan Mode Model (e.g. anthropic/claude-3-7-sonnet with thinking budget).
3. Select your Agent Mode Model (e.g. anthropic/claude-3-5-sonnet or openai/gpt-4o).
4. Settings persist automatically across workspace sessions.

settings.json

{
  "postqode.apiProvider": "postqode",
  "postqode.planModeApiModelId": "anthropic/claude-3-7-sonnet-20250219",
  "postqode.agentModeApiModelId": "anthropic/claude-3-5-sonnet-20241022",
  "postqode.maxThinkingTokens": 8192,
  "postqode.autoApprovalSettings": {
    "enabled": true,
    "actions": {
      "readFiles": true,
      "editFiles": false,
      "executeCommands": false
    }
  }
}

Status Bar Switching

Click the status bar item at the bottom of VS Code to switch models instantly:

postqode:anthropic/claude-sonnet-5  ▾

---

4. API Configuration Profiles

If you switch frequently between low-cost experimentation, daily feature coding, and complex architectural overhauls, use API Configuration Profiles:

🟢 Profile: Budget / Free

Low cost & fast iterations for documentation, unit test boilerplate, and simple refactoring.

• Provider: OpenRouter Free / Ollama
• Model: Qwen 2.5 Coder Free / Devstral
• Cost: $0.00 / turn

🔵 Profile: Daily Driver

Balanced precision, fast diff generation, and dependable tool accuracy for day-to-day development.

• Provider: PostQode / Anthropic
• Model: Claude 3.5 Sonnet / GPT-4o
• Cost: ~$0.01 - $0.03 / turn

🟣 Profile: Heavy Reasoning

Maximum reasoning power for multi-file system redesigns, difficult race condition bugs, and audits.

• Provider: PostQode / Anthropic
• Model: Claude 3.7 Thinking / DeepSeek R1
• Thinking Budget: 16k - 32k tokens