1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
# A trimmed copy of the model lab's roster shape (docs/models/queue.yaml),
# kept in this repo for tests. Real roster files are never read by tests.
meta:
updated: '2026-09-19'
schema: 1
hardware: Apple M5 Max 128 GB (~96 GB comfortable / ~112 GB aggressive)
roster:
- id: coder-daily
role: daily coding agent
name: Qwen3-Coder-Next-80B-A3B
repo: example-org/Qwen3-Coder-Next-4bit
quant: 4bit
ram_gb: 55
backend: mlx_lm
port: 8081
serve: bin/serve example-org/Qwen3-Coder-Next-4bit --port 8081 --host 0.0.0.0 --temp 1.0
grok: local-coder
why: Live. Fixture text.
status: live
- id: coder-next
role: daily coding agent (alias of coder-daily)
name: Qwen3-Coder-Next-80B-A3B
repo: example-org/Qwen3-Coder-Next-4bit
quant: 4bit
ram_gb: 55
backend: mlx_lm
port: 8081
serve: bin/serve example-org/Qwen3-Coder-Next-4bit --port 8081
grok: local-coder-next
why: Same process as coder-daily.
status: live
- id: coder-compact
role: compact fallback coding agent
name: Qwen3-Coder-30B-A3B-Instruct
repo: example-org/Qwen3-Coder-30B-A3B-Instruct-8bit
quant: 8bit
ram_gb: 38
backend: ollama
port: 11434
serve: ollama (tag example-coder:latest)
status: ready
- id: coder-small
role: small best-per-GB coding agent
name: Nemotron-9B-OpenCode
repo: example-org/Nemotron-9B-OpenCode-8bit
quant: 8bit
ram_gb: 9.7
backend: mlx_lm
port: 8092
serve: bin/serve example-org/Nemotron-9B-OpenCode-8bit --port 8092
status: ready
- id: planner
role: planner
name: Qwen3-Next-80B-A3B-Instruct
repo: example-org/Qwen3-Next-80B-A3B-Instruct-4bit
quant: 4bit
ram_gb: 55
backend: mlx_lm
port: 8082
serve: bin/serve example-org/Qwen3-Next-80B-A3B-Instruct-4bit --port 8082
status: ready
- id: swarm-worker
role: swarm worker
name: granite-4.0-h-tiny
repo: example-org/granite-4.0-h-tiny-8bit
quant: 8bit
ram_gb: 7.5
backend: ollama or llamacpp
port: 8084
serve: bin/serve example-org/granite-4.0-h-tiny-8bit --port 8084
status: ready
- id: swarm-micro
role: micro worker
name: Llama-3.2-1B-Instruct
repo: example-org/Llama-3.2-1B-Instruct-8bit
quant: 8bit
ram_gb: 1.4
backend: mlx_lm (chat) or llamacpp (tools)
port: 8085
serve: bin/serve example-org/Llama-3.2-1B-Instruct-8bit --port 8085
status: ready
- id: vlm
role: vision
name: Qwen3-VL-8B
repo: example-org/Qwen3-VL-8B-Instruct-8bit
quant: 8bit
ram_gb: ~10
backend: mlx_vlm
serve: bin/vlm example-org/Qwen3-VL-8B-Instruct-8bit
status: unmeasured
- id: tts
role: voice synthesis
name: Kokoro-82M
repo: example-org/Kokoro-82M-bf16
quant: bf16
ram_gb: 0.33
backend: mlx-audio
serve: bin/tts --voice af_heart
status: ready
- id: voice-llm
role: realtime voice reply model
name: Voice-Small-3B
repo: example-org/Voice-Small-3B-4bit
quant: 4bit
ram_gb: 2
backend: llamacpp
port: 8086
serve: bin/serve-llamacpp voice.gguf --port 8086
status: ready
- id: broken-mem
role: no memory figure
name: Mystery
repo: example-org/Mystery
quant: 4bit
backend: mlx_lm
port: 8087
serve: bin/serve example-org/Mystery
status: ready
- id: "bad id!"
role: invalid token
name: Whatever
repo: example-org/Whatever
ram_gb: 1
backend: mlx_lm
port: 8088
serve: x
status: ready
items:
inbox: