GitHubDocsCocode Nut
Checking sign-in status
01MODEL SERVICE / API 05

A ready-to-use DeepSeek API entry point for your coding agent

Cocode Nut is Cocode’s model-service entry point: sign up once, use a shared balance from the desktop and terminal, and keep the model choice close to the coding workflow.

View plans and pricing↗Explore the Cocode workspace

Capabilities and limits follow the current release

02Direct answer

Start with what it is.

Cocode Nut exposes an OpenAI-compatible model entry point for coding agents, with public support for `/v1/chat/completions` and `/v1/responses`. Model name, protocol, streaming, and tool calling must be verified separately. Legacy `/v1/completions` is not advertised as a universal capability.

Base URL
https://cocode.agency/v1
Chat protocol
/v1/chat/completions
Responses protocol
/v1/responses
Legacy Completions
Not advertised as a universal capability
01

No separate key application

Sign up, sign in, and choose Nut as the model service from Cocode.

02

One balance across clients

The GUI and TUI draw from the same account balance.

03

Built around coding workflows

Nut exposes DeepSeek V4 Pro, V4 Flash, and V4 Flash Vision Exp entry points; availability follows the current plan page.

02Why it matters

Put the model service inside the workflow

Nut is not a separate chat destination. It is a model entry point the Cocode GUI and TUI can select directly, so you do not have to copy API settings between tools.

↗

Clear boundaries make real development work easier to trust.

Bring your own endpoint when you want

If you already have a DeepSeek-compatible API key, you can keep using it. Cocode does not require a single model service.

↗

Clear boundaries make real development work easier to trust.

Pricing, balance, and privacy are explicit

The plans page explains price, credits, model availability, and common questions. Prompts, code, and model responses are not used for model training; necessary metering and request metadata may still be retained.

↗

Clear boundaries make real development work easier to trust.

05Working flow

Move from setup to verification with a clear result at every step.

This is an executable, reviewable path through a real repository—not an abstract feature list.

  1. 01

    Create an account and choose a plan

    Confirm pricing, credits, concurrency, and currently available models on the plan page.

  2. 02

    Obtain API credentials

    Create or use an account credential and store it only in a controlled environment.

  3. 03

    Choose a protocol and model

    Use `/v1/chat/completions` or `/v1/responses`; the selected model must support that protocol.

  4. 04

    Validate with a small request

    Test identity, model availability, and streaming behavior before connecting an agent or automation.

06Use cases

Decide by the work you actually need to finish.

01

Cocode GUI and TUI

Use one signed-in account balance without duplicating configuration across clients.

02

OpenAI-compatible clients

Clients that support the relevant protocol and a custom Base URL can follow current account connection details.

03

Team gateways and internal tools

Connect the model service to a controlled development workflow with explicit credential, quota, and logging boundaries.

07Before you start

Runtime requirements and compatibility boundaries

  • API Base URL: `https://cocode.agency/v1`
  • Chat Completions and Responses are supported; do not assume the legacy `/v1/completions` route exists
  • The model and protocol must both be available; check the account model list and current plans
  • Keep API keys in server-side environment variables or secure credential storage
[ ! ]

Billing, data, and capability are separate boundaries

Plans define quota and priority; privacy terms define data handling; model bindings define protocols and features. Code and model responses are not used for training, while necessary metering, abuse-prevention, and diagnostic metadata may be retained.

08Try it

Responses API request

Confirm model and protocol capability through the account model list, and never commit a real API key.

cocode://workspaceREADY
curl https://cocode.agency/v1/responses \
+  -H "Authorization: Bearer $COCODE_CLOUD_API_KEY" \
+  -H "Content-Type: application/json" \
+  -d '{"model":"deepseek-v4-flash","input":"Review this function for edge cases."}'
09Facts and boundaries

Protocol compatibility must be verified per model and binding.

OpenAI-compatible describes an interface shape. It does not guarantee that every model exposes the same capabilities.

CapabilityStatusConditions and evidence scope
/v1/chat/completionsSupportedThe selected model must also have Chat protocol capability.
/v1/responsesSupportedThe selected model and binding must support the Responses protocol.
/v1/completionsNo universal claimDo not treat legacy Completions as a default capability for every model.
Streaming and toolsConditionalSupport depends on the model, protocol, upstream, and active binding.
10Cocode perspective

API compatibility is not a single boolean.

Cocode tracks the model, protocol, streaming, tool calling, and upstream binding separately. Clients should rely on a capability only when both the request path and model binding support it.

Maintained by
Cocode Agency
Last reviewed
2026-08-25
Scope
Public Cocode Nut API paths, protocols, and model capability boundaries
Verification
Maintained against current routes, model catalog, binding capability, and request-level checks
06FAQ

Before you start, make the boundaries clear.

What is Cocode Nut?+

Cocode Nut is Cocode’s own model-service entry point for calling DeepSeek models from the GUI and TUI.

Is Nut only for Cocode?+

Nut is primarily designed for the Cocode workspace. Supported client access and current connection details follow the account and plans pages.

Is code used for training?+

Cocode’s privacy note says prompts, code, and model responses are not used for model training or sold to third parties; necessary metering and request metadata may be retained.

Does OpenAI-compatible mean every OpenAI SDK feature works?+

No. A client must still verify the target path, model capability, streaming format, tool calling, and error semantics.

Why can a visible model still reject a protocol?+

Catalog visibility and protocol binding are separate capabilities. A listed model is not automatically bound to Chat, Responses, or another route.

07Keep exploring
View Cocode Nut plans↗Self-host a model endpoint↗AI Coding Agent↗Download Cocode↗

A DeepSeek Harness distribution

Product

  • Overview
  • Desktop
  • Terminal
  • Foundation
  • DeepSeek Harness / DSH
  • Local AI coding client
  • Self-hosted coding workflow
  • AI coding agent comparison
  • Download

Use & resources

  • Documentation
  • GitHub repository
  • Cocode Nut
  • Plans & pricing
  • FAQ

Account

  • Sign in
  • Billing & plan

© 2026 cocode agency