Production-ready agentic backend with dynamic system prompts, Ollama integration, GPU scheduling, and MCP tools management
Critical gaps across all dimensions. Tool definitions lack proper schemas, descriptions are minimal or missing, and parameter documentation is sparse. Only 3 of 9 tools have descriptions exceeding 20 characters. No input schemas visible for tools 1, 2, 5, 6, 7. Tool 4 (chat) has a complex schema but lacks output documentation. No error handling guidance, no security annotations, no idempotency markers. This server would not pass code review for production agent use.
Main chat endpoint with dynamic system prompt + options + conversation logging.
Create a persistent custom model (Modelfile-style).
Execute an installed MCP tool with provided parameters.
Health check with proper separate Redis check (not via GPU).
Install a new MCP tool.
List all Ollama models.
List all installed MCP tools.
6 of 9 tools have no visible input schemas (health, list_models, openai_compatible, root, list_tools).
5 of 9 tools have descriptions under 20 characters or missing entirely (health, list_models, openai_compatible, root, list_tools).
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. chat tool returns ChatResponse but structure not visible in definitions.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 35 | <=2025-11-25 | v2 |
OpenAI-compatible endpoint coming in Phase 4
Root endpoint returning service status and tools loaded count.
No error handling guidance. Tools provide no recovery hints (e.g., 'model not found, try list_models first'). Agents cannot self-correct on failures.
Destructive tools (create_model, install_tool, execute_tool) lack confirmation/dry-run patterns and no destructiveHint annotations. Agents cannot safely preview side effects.