SynthoSpeak Orchestrator user manual

SynthoSpeak Orchestrator
user manual

сovers all needs when working with our product - from getting acquainted with it to everyday use and problem‑solving.

📋 Contents

Provider Management Page
https://synthospeak.info/providers

Add Provider
Addition Form

Field	Required	Description
name	✅	Internal name (Latin characters only, no spaces)
Display Name	✅	Displayed name
Type	✅	Provider API type
API Base URL	✅	Provider endpoint URL
API Key	⚠️	Access key (leave empty for Ollama)
Model	✅	Model name
In Quorum	⚠️	Participation in the quorum
Can be Judge	⚠️	Ability to act as a judge

Field: name
Requirements:

Latin letters only
No spaces (use - or _ instead)
Unique within the account
Recommended format: <provider>-<model>

Examples:
✅ mistral-prod
✅ openrouter-nvidia-free
✅ ollama-qwen
❌ Mistral Prod (contains spaces)
❌ мистрал (uses Cyrillic characters)

Field: Display Name
Requirements:

Any characters allowed
Human‑readable name
Displayed in tables and chat

Examples:
✅ Mistral AI
✅ OpenRouter (NVIDIA Free)
✅ Ollama Qwen2.5 0.5B

Field: Type (Dropdown List)

Option	Description	Providers
OpenAI‑compatible	Universal API	OpenAI, Groq, OpenRouter, Together, Mistral, Perplexity
Anthropic (Claude)	Anthropic API	Claude 3, Claude 3.5
xAI (Grok)	xAI API	Grok‑1, Grok‑2
Qwen (Alibaba)	Alibaba Cloud	Qwen, Qwen2, Qwen2.5
DeepSeek	DeepSeek API	DeepSeek‑V3, DeepSeek‑Coder
Ollama (local)	Local models	Any Ollama models
Google Gemini	Google AI	Gemini Pro, Gemini Ultra
YandexGPT	Yandex Cloud	YandexGPT 2, YandexGPT 3

Field: API Base URL
Description: API endpoint URL for requests.
Important: For Ollama, use http://localhost:11434 (without /v1/chat/completions!).

Field: API Key
Description: API access key for the provider.
Exceptions:

Ollama — leave blank (local access)
Some providers — may not require a key for free models

Security:

🔒 Keys are encrypted in the database
🔒 Not displayed in the interface after saving
🔒 Not shared with third parties

Field: Model
Description: The name of the LLM model to be used with the provider.
Requirements:

Must exactly match the model name as specified by the provider
Case‑sensitive in most cases
Should be verified against the provider’s official documentation

Examples:
✅ gpt-4o-mini
✅ claude-3-sonnet
✅ qwen2:7b
✅ gemini-1.5-pro
✅ yandexgpt-lite

Notes:

For Ollama, use the model tag as shown in the ollama list command (e.g., qwen2:0.5b).
For cloud APIs, refer to the provider’s model catalog (e.g., OpenAI’s gpt-4, Anthropic’s claude-3-haiku).
Ensure the model is active and available in your region/plan.

Checkboxes: In Quorum / Can be Judge

Option	Description	Recommendation
In Quorum	The provider participates in response orchestration	✅ For primary models
Can be Judge	The provider can evaluate responses from other models	✅ For powerful models

Recommendations:

Enable In Quorum for all operational providers.
Enable Can be Judge only for powerful models (GPT‑4, Claude, etc.).
For local models (Ollama), only In Quorum is recommended.

Button: Add

Parameter	Value
Action	Save the provider and run an auto‑test
Wait time	5–120 seconds (depends on the provider)
Result	Status: Working or Error

Addition process:
1. Click [Add]
2. Save to the database
3. Automatic connection test (⏳ Testing...)
4. Display the result:
├─ ✅ Working (XXX ms) — success
└─ ❌ Error: ... — error
5. The status is preserved during navigation

Active Providers (Providers Table)
Table Columns

Column	Description
Name	Internal provider name
Model	Model in use
Type	API type (openai, ollama, etc.)
Status	Connection status
Quorum	Participation in the quorum (checkbox)
Judge	Ability to act as a judge (checkbox)
Actions	Action buttons (Test, Delete)

Provider Statuses

Status	Icon	Color	Description
Not Tested	⚪	Gray	Test has not been performed yet
Testing…	🟡	Yellow	Test in progress
Working (XXX ms)	🟢	Green	Connection successful
Error: …	🔴	Red	Connection error

Button: Test (🔄)

Parameter	Value
Color	Blue (#6366F1)
Action	Force connection test
Time	5–120 seconds
Result	Status update

When to use:

After changing the API key
After restoring the connection
To check provider availability

Button: Delete (🗑️)

Parameter	Value
Color	Red (#EF4444)
Action	Delete the provider
Confirmation	Confirmation required
Reversibility	❌ Irreversible

Warning: Deleting a provider cannot be undone!

OpenAI
Obtaining an API Key
https://platform.openai.com/api-keys

Steps:

Log in to your OpenAI account.
Go to API Keys.
Click [Create new secret key].
Copy the key (starts with sk-...).
Save it in a secure location!