AI Cockpit

CLAUDE

CLAUDEArch

0 / ~44K tok

GEMINI

Auto

0 calls · ~0 tok

OLLAMA

qwen2.5-coder:3b

qwen2.5-coder:3b ▼

★ Tier 1 — Best for CRUD & Context

qwen2.5-coder:3b Fast coder, fits VRAM fully, best speed/quality ratio ★ default

qwen2.5-coder:7b Best all-rounder, strong instruction & code tasks ★ quality

deepseek-r1-tool-calling:7b Fine-tuned for structured tool/function calls ★ tools

Tier 2 — Good General Assistants

gemma3:12b Strong reasoning, runs on RAM, good context understanding

gemma3:4b Google model, good reasoning, fits VRAM

llama3.2:1b Lightweight Meta model, decent instruction following

llama3.2:3b Solid Meta model, good instruction following, VRAM-friendly

mistral:7b Strong instruction following, great context retention

qwen2.5-coder:1.5b Very fast coder, minimal VRAM, decent format compliance

qwen2.5:1.5b General Qwen, fast, modest context retention

qwen2.5:3b General Qwen assistant, use coder variant for file tasks

Tier 3 — Reasoning / Slow / Specialized

deepseek-r1:1.5b Reasoning model, wastes tokens on think-blocks for CRUD slow

deepseek-r1-tool-calling:14b Best tool-call quality but runs on RAM, slow interactive slow

gemma2:2b Older Google model, outclassed by gemma3 variants

gemma3:1b Too small for reliable instruction following tiny

gemma4:e2b Multimodal variant, runs on RAM, no text-CRUD advantage RAM

gemma4:e4b Multimodal variant, runs on RAM, no text-CRUD advantage RAM

olmo-3:7b Allen AI, decent chat but inconsistent format compliance

phi Older Microsoft model, mostly a demo, limited reliability

phi3.5 Microsoft model, reasonable but inconsistent on strict output

qwen3:0.6b Thinking model, too small for structured command syntax tiny

qwen3:1.7b Thinking model, better for reasoning than file operations

smollm2:135m Embedded/benchmark use only, not suitable for chat tasks tiny

smollm2:1.7b On-device model, too small for reliable format compliance tiny

smollm2:360m Embedded/benchmark use only, not suitable for chat tasks tiny

stablelm-zephyr:3b Decent chat but weak on structured output formats

starcoder2:3b Code completion model — not instruction/chat, wrong tool

starcoder2:7b Code completion model — not instruction/chat, wrong tool

tinyllama:1.1b Benchmark/toy size, not suitable for real tasks tiny

Embedding Only — No Chat

mxbai-embed-large Text-to-vector embedding for RAG pipelines, not a chat model embed

nomic-embed-text Text-to-vector embedding for RAG pipelines, not a chat model embed

0 calls · 0s ~0 tok

WORKERS

— configure —

0 calls · 0 tok — configure —

TOOLS

— no tools detected —

0 tools no presets

cloud Workers — Online LLM Providers

build Tools — Local CLI Programs

play_arrow Run tool

Cockpit Quick Reference

Keyboard

Files tab

Right panels

Approval drawer

CRUD %% commands

Tips

restart_alt Restart agent sessions?

palette Theme Editor