ChatGPTClaudeGrok / X / CursorGemini

Why DeepSeek Harness Provider IDs Are Permanent

Configuration & UsagePublished 2026-11-18Author: DeepSeek Plugin Market
DeepSeek HarnessDSH plugincustom model APIProvider IDAPI protocol
DeepSeek Harness Provider IDs are permanent because requests, sessions, defaults, and credentials use them. Learn the four references and protocol pairing.

A DeepSeek Harness custom Provider ID cannot be renamed because it is not a label; it is a key used by requests, saved sessions, model defaults, and credential references. The docs say the ID is permanent for exactly those four reasons, while the display name, base URL, protocol, credentials, and models remain editable (source). Understand the references and rename, migration, and dual-protocol gateways all have a correct order.

DeepSeek Harness: Provider ID is a key, not a display name

A Provider ID is data; a display name is a label. Data keys cannot change freely once referenced; labels can change without affecting requests. (source)

First-time custom provider setup often treats the ID as a name, fills in something convenient, and later wants a cleaner value. The field refuses because a rename would break existing requests, sessions, and credential references.

The editable surface is explicit: display name, base URL, protocol, credentials, and models can change; the Provider ID cannot. Decide the ID up front; it is cheaper than migrating later.

A practical rule: use a lowercase, stable ID tied to the endpoint's owner, such as a company gateway or service name. Avoid “test” or “my-api,” which you will want to rename.

DeepSeek Harness: Four references

The ID is used in four places, which is why it cannot change. The official list is requests, saved sessions, model defaults, and credential references (source).

What each reference means:

  1. Requests — every model request resolves an adapter and config by Provider ID; a rename breaks the old route.
  2. Saved sessions — a session log records the provider and model used; a rename can leave it unable to find the provider.
  3. Model defaults — the default model references the Provider ID; a rename leaves a dangling default and blocks input.
  4. Credential references — credentials are associated by provider; a rename can orphan the saved key.

A rename is therefore a migration across requests, sessions, and credentials, not a field edit. Disallowing it trades one restriction for consistency in four places.

DeepSeek Harness: Rename by add, migrate, delete

The correct rename is “add, migrate, delete,” not a hidden edit button. The docs recommend adding a new provider and deleting the old one (source).

Order:

  1. Add the new provider — use the target ID, same base URL, protocol, credentials, and models.
  2. Migrate the default — point the default model at the new provider.
  3. Migrate sessions — switch sessions that must continue, or accept that old sessions keep the old reference.
  4. Verify a request — send a short task with the new provider.
  5. Delete the old provider — only after no default or active session depends on it.

Do not reverse the order. Delete first leaves a dangling default and broken sessions; add first but skip the default leaves input pointing at the old ID. Verify before delete to keep a rollback path.

If the old provider appears in only one test session, migration is cheap. If it is referenced by many sessions and credentials, list the references before touching anything.

DeepSeek Harness: Four protocols

Choose the protocol the gateway actually speaks, not the newest one. The four options are OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, and Gemini API, stored as openai-completions, openai-responses, anthropic-messages, and google-generative-ai (source).

ProtocolConfig valueTypical case
OpenAI Chat Completionsopenai-completionsMost OpenAI-compatible gateways
OpenAI Responsesopenai-responsesEndpoints using the Responses shape
Anthropic Messagesanthropic-messagesClaude-style gateways
Gemini APIgoogle-generative-aiGemini endpoints; URL includes a version

The table tells you which config value to use, not which protocol an endpoint supports. Use the gateway docs and an actual request for that.

A wrong protocol typically looks like “the key and URL are right, but every request is rejected.” Do not rotate keys first; check whether the protocol matches.

DeepSeek Harness: Base URL must match the protocol

A base URL is not any reachable address; it must pair with the protocol. The docs specifically note that Gemini API base URLs include an API version, such as https://gateway.example/v1beta (source).

Three rules:

  1. Protocol decides the path shape — the same gateway has different paths and fields per protocol.
  2. Gemini URLs include a version — the version segment is part of the address, not an afterthought.
  3. One provider, one protocol — a dual-protocol gateway needs two providers with matching URLs.

A missing version, an extra path, or a protocol mismatch usually shows up as 404, 400, or “every request rejected.” It looks like a key problem but is an address-protocol problem.

Model probing is a convenience, not a guarantee. If probing fails or returns an empty list, add model IDs manually; the result is the same (source).

DeepSeek Harness: Verify four references

After migration, verify the four references, not just one request. Four checks (source):

  1. Request — a short task through the new provider returns.
  2. Default — the selector’s default points at the new provider.
  3. Sessions — sessions that must continue use the new model; old sessions behave as expected.
  4. Credential — deleting the old provider leaves requests working.

Only testing the request can miss a stale default; only testing the default can miss a credential that was never migrated. All four make the migration complete.

Two neighboring articles cover catalog refresh and session-scoped model switching. The full custom provider setup is in DeepSeek Harness model providers.

A permanent key is not a limitation to work around; it is the reason four references stay consistent. Pick it once, document it, and change the label when the name changes.

If the ID already exists in several profiles, migrate one profile at a time. A global rename attempt is exactly the operation the design prevents.

Stable keys, editable labels: that is the whole contract.

ChatGPTClaudeGrok / X / CursorGemini

FAQ

Why can I not rename a DeepSeek Harness custom Provider ID?

DeepSeek Harness uses the Provider ID as a permanent key in requests, saved sessions, model defaults, and credential references. Renaming would break those references, so the correct path is add new, migrate, delete old.

What is the difference between a DeepSeek Harness Provider ID and a display name?

In DeepSeek Harness, the Provider ID is a permanent identifier used by requests and sessions; the display name is only a label. DeepSeek Harness lets you edit the name, base URL, protocol, credentials, and models, not the ID.

Which API protocols can a DeepSeek Harness custom model use?

DeepSeek Harness supports OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, and Gemini API, stored as openai-completions, openai-responses, anthropic-messages, and google-generative-ai.

If one gateway supports two protocols, how many DeepSeek Harness providers do I need?

In DeepSeek Harness, two. A DeepSeek Harness provider uses one protocol; a gateway that speaks two needs separate providers, each with a matching base URL.

Related Terms

Provider ID
A Provider ID is the permanent key for a DeepSeek Harness custom model provider, used by requests, sessions, defaults, and credential references.— DeepSeek Harness docs - Configuring models
API protocol
An API protocol is the request shape DeepSeek Harness uses with a model gateway, chosen from four supported options.— DeepSeek Harness docs - Configuring models
base URL
A base URL is the endpoint root DeepSeek Harness sends model requests to; it must match the selected protocol, and Gemini URLs include an API version.— DeepSeek Harness docs - Configuring models
model probing
Model probing is a DeepSeek Harness convenience that asks a custom endpoint for its model list; if it fails, manual model IDs work the same.— DeepSeek Harness docs - Configuring models

Sources