Model provider overview
Provider choice determines where OCR text is processed, how credentials are managed, which models are visible, and which cost modes are available. Validate quality with representative documents before enabling Automatic writes. See Privacy and security before sending real document text to a hosted service.
| Provider | Best for | Authentication / endpoint | Notes |
|---|---|---|---|
| OpenRouter | Curated cloud choice and easy model switching | API key | Recommended hosted starting point; requests are forwarded to the selected upstream provider. |
| Ollama | Fully local inference | Local /api/chat endpoint | Keeps processing on infrastructure you control; model quality and speed depend on hardware. |
| Ollama Cloud | Hosted Ollama models | API key | No local GPU required; document text leaves your network. |
| OpenAI direct | Native OpenAI models, Flex, and Batch | API key | Flex and Batch are available only for supported models. |
| Anthropic direct | Claude and Message Batches | API key | Supports standard requests and discounted asynchronous batches. |
| OpenCode Go | Subscription inference gateway | Go API key | OpenAI-compatible request path with provider-controlled limits. |
| GitHub Copilot | Account-scoped model discovery | OAuth device login or supported token | Uses the official SDK; every agent tool is denied. |
| OpenAI-compatible | LM Studio, LiteLLM, vLLM, custom gateways | Base URL and optional key | Use an endpoint that implements OpenAI Chat Completions. |
| Azure OpenAI | Existing Azure deployments and governance | Endpoint, deployment, API key | Model availability follows your Azure deployment. |
| ChatGPT subscription | Experimental private, low-volume use | Codex device login | Uses the official Codex runtime and the account's live catalog, which may include GPT-5.6 Luna; this is not an API SLA. |
Cost-conscious recommendations
These are starting points for document classification, where consistent structured output matters more than maximum general-purpose reasoning. Test at least 20–50 representative documents in Review first before deciding that a model is good enough for Automatic mode.
| Provider | Recommended starting point | Why it is a good-value choice |
|---|---|---|
| OpenRouter | GPT-5.4 Mini with low reasoning | Best general hosted default in Tagvico: reliable structured extraction without paying for a frontier-sized model. Try GPT-5.4 Nano or Gemini 3.1 Flash Lite for very clean, repetitive documents; avoid the free router for unattended production because the underlying model can change. |
| Ollama | Qwen 3.5 4B on modest hardware; Qwen 3.5 9B when it fits comfortably | Qwen 3.5 is the current value-oriented starting family for structured, multilingual filing. The 4B download is about 3.4 GB; the 9B build is about 6.6 GB and is the better quality target when memory allows. Gemma 3 4B is a strong compact alternative. Gemma 4 is newer, but even its E2B/E4B edge variants have larger model files and are less attractive for a cheapest-first setup. |
| Ollama Cloud | gpt-oss:20b-cloud | Tagvico's default balances capability with a moderate hosted footprint and avoids buying or running a GPU. Recheck the cloud catalog and account limits before committing to it. |
| OpenAI direct | GPT-5.4 Mini; use Batch for non-urgent archives | Mini is the balanced default. GPT-5.4 Nano can reduce cost further for predictable invoices and statements. Batch is preferable when turnaround can wait; Flex is useful when supported and occasional slower availability is acceptable. |
| Anthropic direct | Claude Haiku 4.5; use Message Batches when latency is unimportant | Haiku is the speed-and-cost tier and is normally sufficient for titles, tags, and other structured fields. Move to a larger Claude model only for difficult layouts or extraction failures. |
| OpenCode Go | DeepSeek V4 Flash | This is Tagvico's budget-oriented default for the Go gateway. It suits classification-heavy workloads; confirm the current subscription allowance and gateway model catalog. |
| GitHub Copilot | GPT-5.4 Mini when the signed-in plan exposes it | It offers a strong quality/cost balance without a separate per-token key inside Tagvico. Prefer a model with the lowest billing multiplier that still passes your test set, because plan entitlements differ. |
| OpenAI-compatible | A mini, flash, or roughly 8B–20B instruct model supported by your gateway | Compatible endpoints vary too much for one universal slug. Start small, require reliable JSON/structured output, and increase model size only when the error rate justifies the extra compute or gateway cost. |
| Azure OpenAI | A deployment of GPT-5.4 Mini | Mini is the normal value choice when Azure governance is required. Azure deployment availability and regional pricing take precedence over the public model name. |
| ChatGPT subscription | GPT-5.6 Luna when it appears in the account catalog | Luna is the preferred low-cost GPT-5.6 tier for repetitive filing. Use GPT-5.4 Mini as the fallback. Subscription access is experimental, intended for one trusted low-volume installation, and is not an API service guarantee. |
A practical selection rule
Start with the recommended mini, flash, Haiku, or Luna tier and low reasoning. Measure incorrect or missing fields—not how impressive the prose sounds. Move up one tier only when the cheaper model fails the same field or document type repeatedly.
Ollama sizing notes
The official Qwen 3.5 library currently offers 0.8b, 2b, 4b, 9b, and larger variants—there is no official 7B or 8B tag. For Tagvico, start with:
qwen3.5:4bfor a low-memory trial and routine, clean documents.qwen3.5:9bfor the preferred local balance when the roughly 6.6 GB model file plus runtime overhead fits comfortably.gemma3:4bfor a compact multilingual alternative with a roughly 3.3 GB model file.gemma4:e2borgemma4:e4bonly when you specifically want Gemma 4 and can accommodate their roughly 7.2 GB or 9.6 GB model files. Their “E” sizes mean effective parameters, not download or runtime memory.
Actual RAM or VRAM use is higher than the model file and rises with context length. Keep the context window only as large as your documents require, then compare field accuracy and throughput on the same test set.
Switching providers
Use Settings → AI provider, complete the selected provider form, and test the connection. With environment configuration, change AI_PROVIDER, keep the provider-specific values in data/.env, and restart the container. Each adapter has its own configuration namespace, so switching does not require deleting the previous provider's values.
Processing modes
- Standard processes each document immediately.
- OpenAI Flex reduces cost for supported OpenAI models in exchange for slower or less predictable availability.
- Batch submits asynchronous discounted work to OpenAI or Anthropic and may take up to 24 hours.
Catalogs change
Model names, pricing, quotas, subscription entitlements, and regional availability are provider-controlled. Recheck the provider catalog before each Tagvico release; it is not part of the compatibility contract.
For current model identifiers and account entitlements, use the catalog shown inside Tagvico after authentication. The documentation intentionally does not hard-code model names that providers may rename or withdraw.