> ## Documentation Index
> Fetch the complete documentation index at: https://praison.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# praisonai ollama
> Run local Ollama models from the CLI with Weak-Model-Proof execution
Run a fully local agent against an Ollama model, with Weak-Model-Proof (WMP) execution added automatically to keep small models reliable.
```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
graph LR
subgraph "praisonai ollama"
User[👤 You] --> CLI[💻 praisonai ollama]
CLI --> WMP[🛡️ WMP Wrapper]
WMP --> Server[🧠 Local Ollama]
Server --> Result[✅ Answer]
end
classDef user fill:#8B0000,stroke:#7C90A0,color:#fff
classDef process fill:#189AB4,stroke:#7C90A0,color:#fff
classDef wmp fill:#F59E0B,stroke:#7C90A0,color:#fff
classDef result fill:#10B981,stroke:#7C90A0,color:#fff
class User,CLI user
class Server process
class WMP wmp
class Result result
```
## Quick Start
Ask a question with the default model (`llama3.2:3b`) and auto-enabled WMP.
```bash theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
praisonai ollama "Explain vectors"
```
Point at any model you have pulled locally.
```bash theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
praisonai ollama "Summarise this repo" --ollama-model qwen2.5:7b
```
Turn WMP off for benchmarking raw output or when the model is already reliable.
```bash theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
praisonai ollama "Explain vectors" --no-wmp
```
***
## How It Works
The CLI wraps each request in WMP hooks that validate and retry weak-model output before returning the answer.
```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
sequenceDiagram
participant User
participant CLI as praisonai ollama
participant Handler as OllamaHandler
participant WMP as WMP Hooks
participant Ollama as Local Ollama
User->>CLI: praisonai ollama "task"
CLI->>Handler: build handler + is_available()
Handler->>WMP: auto-config for model size
WMP->>Ollama: enhanced prompt
Ollama-->>WMP: raw response
WMP->>WMP: validate + retry if weak
WMP-->>Handler: verified answer
Handler-->>User: result
```
The handler detects model size from the name and selects a matching WMP profile:
| Model name contains | Detected size | Profile |
| ------------------------ | ------------- | ------------- |
| `70b`, `72b`, `65b` | `large` | fewer retries |
| `13b`, `14b`, `7b`, `8b` | `medium` | balanced |
| anything else | `small` | more retries |
***
## Choosing WMP mode
Match the WMP flag to model size and task complexity.
```mermaid theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
graph TB
Start{Model size?} -->|Large 70B+| Fast[--wmp-fast]
Start -->|Medium 7-14B| Complex{Task complexity?}
Start -->|Small under 7B| Strict[--wmp-strict]
Complex -->|Simple| Fast
Complex -->|Complex| Default[default WMP]
classDef decision fill:#6366F1,stroke:#7C90A0,color:#fff
classDef fast fill:#10B981,stroke:#7C90A0,color:#fff
classDef default fill:#189AB4,stroke:#7C90A0,color:#fff
classDef strict fill:#F59E0B,stroke:#7C90A0,color:#fff
class Start,Complex decision
class Fast fast
class Default default
class Strict strict
```
***
## Command options
Every flag comes from `add_ollama_arguments` in the SDK.
| Flag | Type | Default | Description |
| ------------------------------------ | ----- | -------------------------------------------- | --------------------------------------------------------------- |
| `--provider {ollama,litellm,openai}` | `str` | — | LLM provider to use |
| `--ollama-model` | `str` | `llama3.2:3b` | Ollama model name (e.g. `llama3.2:3b`, `mistral`, `qwen2.5:7b`) |
| `--ollama-host` | `str` | `$OLLAMA_HOST` else `http://localhost:11434` | Ollama server host |
| `--weak-model-proof` / `--wmp` | flag | auto for Ollama | Enable Weak-Model-Proof execution |
| `--no-wmp` | flag | — | Disable Weak-Model-Proof execution |
| `--wmp-strict` | flag | `False` | Stricter validation, more retries |
| `--wmp-fast` | flag | `False` | Fewer retries, faster execution |
| `--step-budget` | `int` | `10` | Maximum steps for WMP execution |
***
## Environment variables
Local Ollama needs no API key; only the server URL matters.
| Variable | Default | Description |
| ------------- | ------------------------ | -------------------------------------------------------- |
| `OLLAMA_HOST` | `http://localhost:11434` | Ollama server URL used when `--ollama-host` is not given |
`--ollama-host` always overrides `OLLAMA_HOST`:
```bash theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
praisonai ollama "Explain vectors" --ollama-host http://192.168.1.10:11434
```
***
## Common patterns
Fast local chat on a capable model:
```bash theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
praisonai ollama "Draft a release note for v2.0" \
--ollama-model qwen2.5:7b \
--wmp-fast
```
Give a small model more room to work through a harder task:
```bash theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
praisonai ollama "Plan a 3-step migration" \
--ollama-model llama3.2:3b \
--wmp-strict \
--step-budget 20
```
Use the same local model inside an agent from Python:
```python theme={"theme":{"light":"vitesse-light","dark":"vitesse-dark"}}
from praisonaiagents import Agent
agent = Agent(
name="Local Agent",
instructions="Answer using the local Ollama model",
llm="ollama/llama3.2"
)
agent.start("Explain vectors")
```
***
## Best Practices
Set `--ollama-model` explicitly so runs are reproducible instead of relying on the default `llama3.2:3b`.
Start `ollama serve` (or the desktop app) before running the command — the CLI exits with a "Make sure Ollama is running" hint if the server is unreachable.
Larger models rarely need many retries, so `--wmp-fast` cuts latency without hurting reliability.
The default of `10` steps handles most tasks; increase it only for genuinely multi-step work to avoid wasted local compute.
***
## Related
Run agents against Ollama with `llm="ollama/…"`, env vars, and YAML.
Run long agent jobs in the background.