speculum.example.toml

# Speculum example config — copy to speculum.toml (repo root or
# ~/.config/speculum/speculum.toml) and edit. The file is optional:
# with none, Speculum probes localhost and shows what it finds.

[server]
host = "127.0.0.1"
port = 8792

[discovery]
enabled = true              # probe the default ports on localhost
# Extra hosts to probe on the same port list — for a collector running away
# from the GPU box (a Proxmox LXC, say). Localhost is always probed and is
# tried first, so a local engine always wins its own port. Claims are keyed on
# host:port, so a remote 8080 stays discoverable while a local 8080 is taken.
# targets = ["10.0.0.41", "gpu-box.lan"]
# Extra ports to probe on localhost and every target, for an engine running on
# a non-default port. Each new port is tried with every adapter, openai last.
# ports = [8081, 9091]

# The built-in threads (llama-swap, and the NInfer backend that follows its
# /running proxy) are not discovery. Move them off their stock ports with a
# `port`, a `host`, or a full `url`.
# [llama_swap]
# port = 9091                # default http://127.0.0.1:9090
#
# [strata]
# url = "http://127.0.0.1:8090"   # default http://127.0.0.1:8080

[history]
retention_days = 30         # 30, 60, 90 or 0 to turn history off

# Any number of engines, on any host. Type names the adapter; the URL
# may point anywhere (the collector is just an HTTP client). `port` and
# `host` stand in for a full url, so an engine off its stock port is a
# one-line change.

[[engine]]
name = "ollama-desk"
type = "ollama"
url  = "http://10.0.0.41:11434"

# llama.cpp on a custom port: no URL needed, host defaults to 127.0.0.1.
# [[engine]]
# name = "llama.cpp"
# type = "llamacpp"
# port = 8081

# llama-swap, NInfer and Strata are already watched by the built-in
# threads; adding them here again is not needed (phase 1b folds those
# threads into adapters).

[[engine]]
name = "my-engine"
type = "custom"
url  = "http://127.0.0.1:7000"
health  = "/health"
models  = "/v1/models"
metrics = "/metrics"        # Prometheus text
[engine.map]                # schema field = metric name in the engine's /metrics
output_tokens = "myengine_generated_tokens_total"
prompt_tokens = "myengine_prompt_tokens_total"
queue         = "myengine_requests_waiting"

# Apps you start now and then: `optional = true` keeps a card that says "not running" and raises
# no alert while they are down.
#
# [[engine]]
# name     = "LM Studio"
# type     = "lmstudio"
# url      = "http://127.0.0.1:1234"
# optional = true
#
# Unsloth Studio (`unsloth studio`, default port 8888). /api/health needs no key; listing its models
# does: make an API key in Studio's settings and put it in UNSLOTH_STUDIO_AUTH_TOKEN (the default).
#
# [[engine]]
# name        = "Unsloth"
# type        = "unsloth"
# url         = "http://127.0.0.1:8888"
# optional    = true
# api_key_env = "UNSLOTH_STUDIO_AUTH_TOKEN"

# Secrets stay out of this file: name the environment variable that
# holds the bearer token and Speculum sends it as Authorization.
#
# [[engine]]
# name = "paywalled"
# type = "openai"
# url  = "https://api.example.com/v1"
# api_key_env = "MY_ENGINE_KEY"