deepseek-ai/awesome-deepseek-agent

Fork 0

mirror of https://github.com/deepseek-ai/awesome-deepseek-agent.git synced 2026-11-05 01:07:37 +00:00

[PR #429] docs: add Grok Build integration guide #427

New issue

Open

opened 2026-10-26 09:49:10 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 09:49:10 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/awesome-deepseek-agent/pull/429
Author: @tiandi02
Created: 9/25/2026
Status: 🔄 Open

Base: main ← Head: docs/add-grok-build

📝 Commits (1)

74606b6 docs: add Grok Build integration guide

📊 Changes

4 files changed (+238 additions, -0 deletions)

View changed files

📝 README.md (+1 -0)
📝 README.zh-CN.md (+1 -0)
➕ docs/grok_build.md (+118 -0)
➕ docs/grok_build.zh-CN.md (+118 -0)

📄 Description

Summary

Adds a bilingual integration guide for Grok Build (grok), xAI's terminal coding agent:

docs/grok_build.md (English) and docs/grok_build.zh-CN.md (Simplified Chinese), in one PR.
README table entries in both README.md and README.zh-CN.md, inserted alphabetically (after GitHub Copilot CLI, before Hermes).

Relationship to the other open Grok Build PRs

As CONTRIBUTING asks, I checked the open PRs first. #314, #318 and #405 all cover this tool and are still open, so I want to be explicit rather than silently add a fourth:

	#314	#318	#405	this PR
Backend	Chat Completions	both	Anthropic `messages` stated as required	Chat Completions (the default) — verified working
Reasoning effort	`supports_reasoning_effort` (deprecated)	`default_reasoning_effort = "high"`	`default_reasoning_effort = "max"`	`reasoning_efforts` + `max`, wire-verified
Premise	—	—	Chat Completions hits a `reasoning_content` 400, so the Anthropic backend is needed	That 400 was not reproducible on Grok 1.0.41

I am happy to close this in favor of any of them, or to fold anything useful from here into one of them, if a maintainer prefers a single Grok Build entry — #405 already made the same offer about #314.

Verified against Grok 1.0.41, not inferred

CONTRIBUTING asks that config fields be verified to actually work in the agent, so each claim below comes from a recorded observation:

Model naming. deepseek-v4-pro and deepseek-v4-flash both work; deepseek-v4-flash normalizes to deepseek-flash in the response model field.
1M context. context_window = 1000000; DeepSeek's /v1/models row reports context_window: 1048576. Grok uses this value to time auto-compaction, which matches the config reference.
Max thinking effort, wire-verified. The request body was captured on a local endpoint. Grok builds the effort menu from the endpoint's /v1/models row, where DeepSeek reports levels as effort.supported_levels — a shape Grok does not read. Consequences, all measured:
- With no reasoning_efforts line, no reasoning field is sent at all, even with default_reasoning_effort = "max" and --effort max.
- With reasoning_efforts = ["low", "high", "max"] and default_reasoning_effort = "max", the body carries "reasoning_effort":"max".
- --effort max overrides a lower configured default.
- Note for the config reference: it describes reasoning_efforts as an array of tables; the array-of-strings form above is the one that takes effect here.
No protocol workaround needed. A headless run on the default chat_completions backend completed a multi-turn task that invoked shell and file tools, with no reasoning_content error. That is why this guide does not tell users to switch backends or to disable thinking.
Credential fallback, confirmed. With env_key naming an unset variable and no api_key, Grok does not report a missing key: the request carried a bearer token whose SHA-256 matches the auth.json OIDC session token, i.e. the user's grok.com session credential is sent to the third-party endpoint. This is documented as a pitfall in both language versions.

Checks

English and Chinese in one PR
README entries in both languages, alphabetical
Current model names (deepseek-v4-pro / deepseek-v4-flash), no v3 names
1M context configured
max reasoning effort supported, verified on the wire
One tool per PR
No workarounds for already-fixed upstream bugs

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/deepseek-ai/awesome-deepseek-agent/pull/429 **Author:** [@tiandi02](https://github.com/tiandi02) **Created:** 9/25/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `docs/add-grok-build` --- ### 📝 Commits (1) - [`74606b6`](https://github.com/deepseek-ai/awesome-deepseek-agent/commit/74606b610abb1f9eb473a60f9668dcaa9da32082) docs: add Grok Build integration guide ### 📊 Changes **4 files changed** (+238 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `README.md` (+1 -0) 📝 `README.zh-CN.md` (+1 -0) ➕ `docs/grok_build.md` (+118 -0) ➕ `docs/grok_build.zh-CN.md` (+118 -0) </details> ### 📄 Description ## Summary Adds a bilingual integration guide for **Grok Build** (`grok`), xAI's terminal coding agent: - `docs/grok_build.md` (English) and `docs/grok_build.zh-CN.md` (Simplified Chinese), in one PR. - README table entries in both `README.md` and `README.zh-CN.md`, inserted alphabetically (after GitHub Copilot CLI, before Hermes). ## Relationship to the other open Grok Build PRs As CONTRIBUTING asks, I checked the open PRs first. #314, #318 and #405 all cover this tool and are still open, so I want to be explicit rather than silently add a fourth: | | #314 | #318 | #405 | this PR | | --- | --- | --- | --- | --- | | Backend | Chat Completions | both | Anthropic `messages` stated as required | **Chat Completions (the default) — verified working** | | Reasoning effort | `supports_reasoning_effort` (deprecated) | `default_reasoning_effort = "high"` | `default_reasoning_effort = "max"` | `reasoning_efforts` + `max`, wire-verified | | Premise | — | — | Chat Completions hits a `reasoning_content` 400, so the Anthropic backend is needed | That 400 was **not** reproducible on Grok 1.0.41 | I am happy to close this in favor of any of them, or to fold anything useful from here into one of them, if a maintainer prefers a single Grok Build entry — #405 already made the same offer about #314. ## Verified against Grok 1.0.41, not inferred CONTRIBUTING asks that config fields be verified to actually work in the agent, so each claim below comes from a recorded observation: - **Model naming.** `deepseek-v4-pro` and `deepseek-v4-flash` both work; `deepseek-v4-flash` normalizes to `deepseek-flash` in the response `model` field. - **1M context.** `context_window = 1000000`; DeepSeek's `/v1/models` row reports `context_window: 1048576`. Grok uses this value to time auto-compaction, which matches the config reference. - **Max thinking effort, wire-verified.** The request body was captured on a local endpoint. Grok builds the effort menu from the endpoint's `/v1/models` row, where DeepSeek reports levels as `effort.supported_levels` — a shape Grok does not read. Consequences, all measured: - With no `reasoning_efforts` line, **no reasoning field is sent at all**, even with `default_reasoning_effort = "max"` *and* `--effort max`. - With `reasoning_efforts = ["low", "high", "max"]` and `default_reasoning_effort = "max"`, the body carries `"reasoning_effort":"max"`. - `--effort max` overrides a lower configured default. - Note for the config reference: it describes `reasoning_efforts` as an *array of tables*; the array-of-strings form above is the one that takes effect here. - **No protocol workaround needed.** A headless run on the default `chat_completions` backend completed a multi-turn task that invoked shell and file tools, with no `reasoning_content` error. That is why this guide does not tell users to switch backends or to disable thinking. - **Credential fallback, confirmed.** With `env_key` naming an unset variable and no `api_key`, Grok does not report a missing key: the request carried a bearer token whose SHA-256 matches the `auth.json` OIDC session token, i.e. the user's grok.com session credential is sent to the third-party endpoint. This is documented as a pitfall in both language versions. ## Checks - [x] English and Chinese in one PR - [x] README entries in both languages, alphabetical - [x] Current model names (`deepseek-v4-pro` / `deepseek-v4-flash`), no v3 names - [x] 1M context configured - [x] `max` reasoning effort supported, verified on the wire - [x] One tool per PR - [x] No workarounds for already-fixed upstream bugs --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi added the

pull-request

label

2026-10-26 09:49:10 +00:00

No labels

pull-request

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

deepseek-ai/awesome-deepseek-agent#427

No description provided.

Rows
Columns