protoResearcher/docker-compose.yml at main · protoLabsAI/protoResearcher · GitHub

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
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
services:
  researcher:
    build:
      context: .
      dockerfile: Dockerfile
    container_name: protoresearcher
    hostname: sandbox
    restart: unless-stopped

    # NemoClaw-inspired security
    security_opt:
      - no-new-privileges:true
      - seccomp=seccomp-profile.json
    cap_drop:
      - ALL
    cap_add:
      - NET_RAW
    read_only: true

    # Writable mounts (tmpfs = ephemeral)
    tmpfs:
      - /tmp:size=512M,uid=1001,gid=1001
      - /run:size=64M
      - /sandbox:size=512M,uid=1001,gid=1001
      - /home/sandbox:size=128M,uid=1001,gid=1001

    # Persistent volumes
    volumes:
      - protoresearcher-knowledge:/sandbox/knowledge
      - protoresearcher-audit:/sandbox/audit
      - protoresearcher-papers:/sandbox/papers
      - protoresearcher-cron:/opt/.cron
      # Config (bind-mount for APO prompt optimization)
      - ./config:/opt/protoresearcher/config:rw

    ports:
      - "7872:7870"

    # Network: host vLLM + Ollama access
    extra_hosts:
      - "host.docker.internal:host-gateway"

    # Resource limits
    deploy:
      resources:
        limits:
          cpus: "4"
          memory: 4G
        reservations:
          memory: 512M

    environment:
      - HOME=/home/sandbox
      - AGENT_BACKEND=${AGENT_BACKEND:-langgraph}
      - OPENAI_API_KEY=${LITELLM_MASTER_KEY:-}
      - LITELLM_MASTER_KEY=${LITELLM_MASTER_KEY:-}
      # Langfuse tracing (optional)
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY:-}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY:-}
      - LANGFUSE_HOST=${LANGFUSE_HOST:-http://host.docker.internal:3001}
      # GitHub (optional, for higher API rate limits)
      - GITHUB_TOKEN=${GITHUB_TOKEN:-}
      # Discord (optional, for reading channel feeds)
      - DISCORD_BOT_TOKEN=${DISCORD_RESEARCH_BOT:-}
      - DISCORD_WEBHOOK_URL=${DISCORD_WEBHOOK_URL:-}
      # Instance identity (for multi-node collaboration)
      - INSTANCE_NAME=${INSTANCE_NAME:-ava}
      # Rabbit Hole integration
      - MCP_AUTH_TOKEN=${MCP_AUTH_TOKEN:-}
      - RABBIT_HOLE_URL=${RABBIT_HOLE_URL:-http://host.docker.internal:3399}
      # A2A authentication — workstacean sends this as X-API-Key
      - RESEARCHER_API_KEY=${RESEARCHER_API_KEY:-}

    stdin_open: true
    tty: true

    networks:
      default:
      gateway:
        aliases:
          - researcher   # workstacean resolves http://researcher:7870/a2a via ai_default

    command: ["/opt/protoresearcher/entrypoint.sh"]

  # Lab mode — GPU-enabled researcher with training capabilities
  # Usage: docker compose --profile lab up --build
  researcher-lab:
    build:
      context: .
      dockerfile: Dockerfile
      target: lab
    container_name: protoresearcher-lab
    hostname: sandbox
    restart: unless-stopped
    profiles:
      - lab

    # Relaxed security for CUDA (GPU needs syscalls seccomp blocks)
    security_opt:
      - no-new-privileges:true
    cap_drop:
      - ALL
    cap_add:
      - NET_RAW

    tmpfs:
      - /tmp:size=1G,uid=1001,gid=1001
      - /run:size=64M
      - /sandbox:size=1G,uid=1001,gid=1001
      - /home/sandbox:size=256M,uid=1001,gid=1001

    volumes:
      - protoresearcher-knowledge:/sandbox/knowledge
      - protoresearcher-audit:/sandbox/audit
      - protoresearcher-papers:/sandbox/papers
      - protoresearcher-lab:/sandbox/lab
      - protoresearcher-cron:/opt/.cron
      # Host mounts for training
      - /mnt/models/huggingface:/mnt/models/huggingface:ro
      - /home/ava/dev/LLaMA-Factory:/opt/llama-factory:ro
      - /home/ava/dev/lab/training/data:/opt/lab-data:ro
      - /mnt/data/training/researcher:/mnt/data/training/researcher
      # Claude OAuth credentials (read-only)
      - ${HOME}/.claude:/opt/claude-creds:ro

    ports:
      - "7870:7870"

    extra_hosts:
      - "host.docker.internal:host-gateway"

    # GPU access
    deploy:
      resources:
        limits:
          cpus: "8"
          memory: 16G
        reservations:
          memory: 2G
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

    environment:
      - HOME=/home/sandbox
      - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY:-}
      - LAB_GPU=${LAB_GPU:-1}
      - HF_HOME=/mnt/models/huggingface
      - LANGFUSE_PUBLIC_KEY=${LANGFUSE_PUBLIC_KEY:-}
      - LANGFUSE_SECRET_KEY=${LANGFUSE_SECRET_KEY:-}
      - LANGFUSE_HOST=${LANGFUSE_HOST:-http://host.docker.internal:3001}
      - GITHUB_TOKEN=${GITHUB_TOKEN:-}
      - DISCORD_BOT_TOKEN=${DISCORD_BOT_TOKEN:-}
      - DISCORD_WEBHOOK_URL=${DISCORD_WEBHOOK_URL:-}
      - INSTANCE_NAME=${INSTANCE_NAME:-protolabs}

    stdin_open: true
    tty: true

    command: ["/opt/protoresearcher/entrypoint.sh"]

networks:
  gateway:
    external: true
    name: ai_default

volumes:
  protoresearcher-knowledge:
  protoresearcher-audit:
  protoresearcher-papers:
  protoresearcher-cron:
  protoresearcher-lab: