Audit meines LLM-Verbrauchs: SQLite, OpenCode, Ollama Cloud und warum mein Pro-Abonnement den globalen Markt sprengt

Failed to generate image: PlantUML preprocessing failed: [From <input> (line 10) ]

@startuml
skinparam backgroundColor #FEFEFE
skinparam defaultFontSize 12

title Die drei Akteure meines LLM-Stacks
left to right direction

actor "Ich (Entwickler)" as Dev

package "OpenCode
^^^^^
 Syntax Error? (Assumed diagram type: component)

@startuml
skinparam backgroundColor #FEFEFE
skinparam defaultFontSize 12

title Die drei Akteure meines LLM-Stacks
left to right direction

actor "Ich (Entwickler)" as Dev

package "OpenCode
(github.com/anomalyco/opencode)" as OC {
  rectangle "CLI von vibe coding
Open Source, kostenlos" as CLI
  note right of CLI
    C'est l'outil que j'utilise
    pour coder avec des agents IA.
    Il ne fournit AUCUN modèle.
    C'est un terminal + un orchestrateur.
  end note
}

package "Ollama (Lokale Installation)" as OL {
  rectangle "Inferenzclient
Auf meinem Rechner installiert" as Client
  note right of Client
    C'est le logiciel qui fait tourner
    les modèles. Il peut les exécuter
    en local OU les déléguer au cloud
    selon le tag (:cloud).
    Il s'authentifie via clé SSH.
  end note
}

package "Ollama Cloud\n(ollama.com)" as OCloud {
  rectangle "Cloud-Inferenz-Dienst
Pro-Abonnement: 20€/Monat" as Cloud
  note right of Cloud
    C'est le fournisseur d'inférence.
    Mes modèles avec le tag :cloud
    sont exécutés sur LEURS serveurs.
    L'abonnement Pro donne accès
    à des quotas supérieurs.
  end note
}

Dev --> CLI : "Ich tippe Prompts"
CLI --> Client : "Rufe dieses Modell"
Client --> Cloud : "SSH-Schlüssel + Anfrage
(Modelle :cloud)"
Client --> Client : "Lokale Ausführung
(Modelle ohne :cloud)"

@enduml

$ ls -lh ~/.local/share/opencode/opencode.db
-rw-r--r-- 1 cheroliv cheroliv 660M mai 18 12:25 opencode.db

Failed to generate image: PlantUML preprocessing failed: [From <input> (line 8) ]

@startuml
skinparam backgroundColor #FEFEFE
skinparam defaultFontSize 11

title Projizierte monatliche Kosten — 220M Eingabe + 1,9M Ausgabe
rectangle "Ollama Cloud Pro\n~20€" as OC #4CAF50
rectangle "Gemini Flash
$69" as GF #8BC34A
^^^^^
 Syntax Error? (Assumed diagram type: activity)

@startuml
skinparam backgroundColor #FEFEFE
skinparam defaultFontSize 11

title Projizierte monatliche Kosten — 220M Eingabe + 1,9M Ausgabe
rectangle "Ollama Cloud Pro\n~20€" as OC #4CAF50
rectangle "Gemini Flash
$69" as GF #8BC34A
rectangle "DeepSeek-Promo
\$98" as DSP #FFC107
rectangle "GPT Mini
$174" as GM #FF9800
rectangle "Claude Haiku\n\$230" as CH #FF5722
rectangle "HF/Novita\n\$374" as HF #F44336
rectangle "DeepSeek normal
\$390" as DSN #F44336
rectangle "GPT 5.4\n\$579" as G4 #D32F2F
rectangle "Claude Sonnet
\$689" as CS #D32F2F
rectangle "Claude Opus
$1 148" as CO #B71C1C
rectangle "GPT 5.5\n\$1 157" as G5 #880E4F

note bottom of OC
  DeepSeek V4 Pro 1.6T MoE
  + accès à tout le catalogue
  Ollama Cloud (Qwen, GLM, Kimi...)
  Inclus dans l'abonnement
end note
@enduml

Failed to generate image: PlantUML preprocessing failed: [From <input> (line 9) ]

@startuml
skinparam backgroundColor #FEFEFE
skinparam defaultFontSize 12

title Architektur — Zwei Ollama Cloud Pro‑Abonnements auf einer Maschine
left to right direction

node "Meine physische Maschine" {
  frame "Ollama Instanz A (einheimisch)
^^^^^
 Syntax Error? (Assumed diagram type: class)

@startuml
skinparam backgroundColor #FEFEFE
skinparam defaultFontSize 12

title Architektur — Zwei Ollama Cloud Pro‑Abonnements auf einer Maschine
left to right direction

node "Meine physische Maschine" {
  frame "Ollama Instanz A (einheimisch)
Port :11434" as Native {
    database "Identität A\n/usr/share/ollama/.ollama/id_ed25519"
  }
  frame "Ollama-Instanz B (Docker)
Port :11435" as Docker {
    database "Identität B\n~/ollama-b-data/id_ed25519"
  }
}

cloud "Ollama Cloud" {
  actor "Konto Pro A\n(E‑Mail 1)" as A
  actor "Geschäftskonto B\n(E-Mail 2)" as B
}

Native --> A : "Gerätenschlüssel A"
Docker --> B : "Geräteschlüssel B"

note bottom of Docker
  Conteneur Docker isolé :
  - Clé SSH distincte
  - Port réseau distinct (11435)
  - Volume persistant (~/ollama-b-data)
  - restart: always
end note
@enduml