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
|
{ config, pkgs, hermes-agent, ... }:
{
imports = [
hermes-agent.nixosModules.default
];
age.secrets.renard-agent = {
file = ../secrets/renard-agent.age;
owner = "renard";
group = "renard";
mode = "0440";
};
users.groups.renard = { };
users.users.renard = {
isSystemUser = true;
group = "renard";
home = "/home/renard";
createHome = true;
shell = pkgs.bashInteractive;
};
services.hermes-agent = {
enable = true;
# the agent has it's own user on the host under
# which it runs and can change state at.
user = "renard";
group = "renard";
stateDir = "/home/renard";
createUser = false;
extraDependencyGroups = [ "messaging" ];
# packages for the agent to work with.
# it should also always be able to call `nix-shell` to gain access to more tools.
extraPackages = with pkgs; [
nix
nixfmt-rfc-style
openssh
tailscale
docker_29
jq
yq-go
ripgrep
fd
tree
file
unzip
gnused
gawk
curl
wget
htop
glances
lsof
iproute2
iputils
inetutils
dig
tcpdump
nmap
git
vim
];
environmentFiles = [
# required secrets:
# * OPENROUTER_API_KEY
# * DISCORD_BOT_TOKEN
# * DISCORD_ALLOWED_USERS
config.age.secrets.renard-agent.path
];
documents = {
"SOUL.md" = ../assets/agent/SOUL.md;
};
settings = {
model = {
provider = "openrouter";
# my current favorite budget model, with recent price
# changes by far the cheapest for the performance,
# with other chinese models also being close.
# only modality is text, requires auxiliary vision model!
# native price: $0.44/m in, $0.87/m out.
default = "deepseek/deepseek-v4-pro";
};
agent = {
reasoning_effort = "high"; # 'high' rated highest in benchmarks, tune to taste!
max_turns = 90;
};
auxiliary.vision = {
# auxiliary vision model for image comprehension.
# this qwen model is built for this exact use-case and is extremely
# cheap for the performance (though below gemini), however,
# it does not do any reasoning on the visual input!
# possible upgrade path for reasoning: qwen3-vl-30b-a3b-thinking
# native price: $0.13/m in, $1.56/m out.
provider = "openrouter";
model = "qwen/qwen3-vl-32b-instruct";
timeout = 120;
};
toolsets = [ "all" ];
terminal = {
backend = "local";
persistent_shell = true;
timeout = 180;
};
memory = {
memory_enabled = true;
user_profile_enabled = true;
};
# slightly higher limits and targets than default for better
# long session preservation.
compression = {
enabled = true;
threshold = 0.75;
target_ratio = 0.35;
protect_last_n = 40;
};
# progressive message editing, a little silly in discord but looks like
# the classic llm chats.
streaming = {
enabled = true;
};
display = {
streaming = true; # small regression with streaming.enabled, keep both on.
tool_progress = "new"; # only show fresh new tool calls, not every call.
};
discord = {
require_mention = false;
thread_require_mention = false;
auto_thread = false;
reactions = true;
history_backfill = true;
allow_mentions = {
everyone = false;
roles = false;
users = true;
replied_user = true;
};
};
unauthorized_dm_behavior = "ignore"; # ignore strangers
timezone = "Europe/Berlin";
group_sessions_per_user = false;
};
};
}
|
