Run an instance with no provider key

Declare no provider on either sink, and this instance reads, links and searches without an LLM. Nothing leaves it, and no key is asked for.

Before you start

• Create an instance with revetly init
• No provider key and no provider account: nothing here reaches a vendor.

Configuration page

1. Open the configuration page. If it is not open, run revetly config and open the address it prints.

   You should see One answer still needed.

2. Choose No provider (lite instance) in LLM provider. Revetly carries the embedding provider into the same save.

   You should see Embedding provider is saved as No provider (lite instance) too. Revetly starts on no provider for both or a provider for both, never one of each.

3. Select Save.

   You should see Configured and running.

4. Read the provider line the page carries now.

   You should see LLM and embedding: No provider (lite instance) on the instance card.

Command line

1. Scaffold an instance that declares no provider:

   revetly init --lite
   

2. In revetly.toml at the project root, declare none on both sinks of an instance that already exists:

   # revetly.toml
   [llm]
   provider = "none"
   
   [embedding]
   provider = "none"
   

   If revetly init --lite wrote those two tables, leave them as they are.

3. Read the whole chain back, before anything is spent:

   revetly doctor --offline
   

   [PASS] lite   lite instance — `[llm] provider = "none"` and `[embedding] provider = "none"`: every kind is indexed, links are deterministic and search is keyword-only …
   [PASS] egress (extraction)   none — no provider is declared, so no body goes anywhere; nothing is extracted and every kind is indexed …
   [PASS] egress (embeddings)   none — no provider is declared, so no body goes anywhere; vectors are the engine's own stand-ins, which nothing measured …
   

   Revetly names no provider key among those lines, because a lite instance reads none.

4. Price the first pass before running it:

   revetly ingest --dry-run
   

   would spend: this instance declares no provider (`[llm] provider = "none"`), so every episode here is written by the engine itself: no body and no query leaves, nothing is extracted, and this pass costs nothing to run
   would buy: zero tokens and zero cost — this instance declares no provider, so nothing here is ever billed in either
   

Troubleshooting

• The declaration is refused and both provider keys are named. One sink declares none and the other names a provider. Declare none under [llm] and under [embedding] together.
• A search comes back with no concepts. That is the declaration working rather than a failure. Name a provider on both sinks to turn extraction on.

Next steps

• Give an instance a provider key and confirm it
• Price and run the first pass, and know it landed