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 #442] docs(ember): add Ember integration guide (EN + zh-CN) #438

New issue

Open

opened 2026-10-26 09:49:23 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 09:49:23 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/awesome-deepseek-agent/pull/442
Author: @yyfdfsyyf-source
Created: 10/6/2026
Status: 🔄 Open

Base: main ← Head: add-ember-guide

📝 Commits (1)

f3d6b07 docs(ember): add Ember integration guide (EN + zh-CN)

📊 Changes

4 files changed (+243 additions, -0 deletions)

View changed files

📝 README.md (+1 -0)
📝 README.zh-CN.md (+1 -0)
➕ docs/ember.md (+127 -0)
➕ docs/ember.zh-CN.md (+114 -0)

📄 Description

Adds an integration guide for Ember, a text-first open-source terminal coding agent written in C++20 (AGPL-3.0-only), for DeepSeek V4.

One PR, both languages: docs/ember.md + docs/ember.zh-CN.md, plus one row inserted in alphabetical order in both README.md and README.zh-CN.md (right after DeepSeek-TUI). No images.

Critical checks

1. Model naming — only deepseek-v4-pro / deepseek-v4-flash appear. grep -n "deepseek-chat\|deepseek-reasoner\|deepseek-coder" docs/ember.md docs/ember.zh-CN.md is empty.

2. 1M context — Ember has no per-model context_window key, so the guide says that out loud and documents the mechanism it does have: budget (default 6000; 0 disables compression), prune_before_send, --tool-cap N, and notes that keeping budget under 1M is what stays inside the limit.

3. Max thinking — "thinking": "max", /think max and --thinking max are all documented; max is a first-class level in auto|none|minimal|low|medium|high|max. DeepSeek's reasoning_effort spelling is Ember's default thinking_style (an unset or unrecognized value normalizes to effort), so connecting DeepSeek needs no extra key. The guide never recommends turning thinking off to dodge an API error.

4. Pricing — deliberately not included, so there is nothing to drift out of date.

5. Config fields are verified to work in the agent — verified by recording Ember's own requests. Ember 0.10.1 was driven through its headless --serve protocol against a local OpenAI-compatible endpoint that dumps every request; two turns:

Turn 1, using exactly the snippet in the guide (providers[0].api_key_env = "DEEPSEEK_API_KEY", group ds):
POST /v1/chat/completions · Authorization: Bearer sk-test-wire-verify-123 (the key came from the environment variable, not from the file) · model: deepseek-v4-pro · reasoning_effort: "max" · stream: true · body keys [messages, model, reasoning_effort, stream, temperature, tools].
Then {"type":"set","key":"model","value":"other-model"}, where that model belongs to a different provider/group. The next request went to that provider's base_url with Authorization: Bearer inline-other-key and model: other-model — so the guide's claim that switching a grouped model moves the base URL and the key in one step is what actually happens on the wire.

Scope, stated plainly: this verifies Ember's config fields and request construction. It was not run against api.deepseek.com — no DeepSeek key is available on the machine I verified from — so the DeepSeek-side behaviour in the guide comes from your own docs (reasoning_effort = low|high|max, V4 1M context) rather than from my own API calls. Happy to add a live-API confirmation as a follow-up commit.

6. No workarounds for already-fixed upstream bugs — none suggested.

Install paths referenced

The pre-built packages link to the v0.10.1 release assets (both verified served; the SHA256 of each is listed in the release notes), and the same bytes are also committed in-tree in the repo. Source builds cover build.ps1 (plain g++), the CMake path (cmake -B build -G Ninja && cmake --build build && ctest — 23/23 passing), and build-linux.sh, which is what produces the released Linux binary.

_{🔄 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/442 **Author:** [@yyfdfsyyf-source](https://github.com/yyfdfsyyf-source) **Created:** 10/6/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `add-ember-guide` --- ### 📝 Commits (1) - [`f3d6b07`](https://github.com/deepseek-ai/awesome-deepseek-agent/commit/f3d6b077f65bf3fc5ba311d6b7bd58bceb32138b) docs(ember): add Ember integration guide (EN + zh-CN) ### 📊 Changes **4 files changed** (+243 additions, -0 deletions) <details> <summary>View changed files</summary> 📝 `README.md` (+1 -0) 📝 `README.zh-CN.md` (+1 -0) ➕ `docs/ember.md` (+127 -0) ➕ `docs/ember.zh-CN.md` (+114 -0) </details> ### 📄 Description Adds an integration guide for **Ember**, a text-first open-source terminal coding agent written in C++20 (AGPL-3.0-only), for DeepSeek V4. One PR, both languages: `docs/ember.md` + `docs/ember.zh-CN.md`, plus one row inserted in alphabetical order in **both** `README.md` and `README.zh-CN.md` (right after DeepSeek-TUI). No images. ## Critical checks **1. Model naming** — only `deepseek-v4-pro` / `deepseek-v4-flash` appear. `grep -n "deepseek-chat\|deepseek-reasoner\|deepseek-coder" docs/ember.md docs/ember.zh-CN.md` is empty. **2. 1M context** — Ember has no per-model `context_window` key, so the guide says that out loud and documents the mechanism it does have: `budget` (default `6000`; `0` disables compression), `prune_before_send`, `--tool-cap N`, and notes that keeping `budget` under 1M is what stays inside the limit. **3. Max thinking** — `"thinking": "max"`, `/think max` and `--thinking max` are all documented; `max` is a first-class level in `auto|none|minimal|low|medium|high|max`. DeepSeek's `reasoning_effort` spelling is Ember's **default** `thinking_style` (an unset or unrecognized value normalizes to `effort`), so connecting DeepSeek needs no extra key. The guide never recommends turning thinking off to dodge an API error. **4. Pricing** — deliberately not included, so there is nothing to drift out of date. **5. Config fields are verified to work in the agent** — verified by recording Ember's own requests. Ember 0.10.1 was driven through its headless `--serve` protocol against a local OpenAI-compatible endpoint that dumps every request; two turns: - Turn 1, using exactly the snippet in the guide (`providers[0].api_key_env = "DEEPSEEK_API_KEY"`, group `ds`): `POST /v1/chat/completions` · `Authorization: Bearer sk-test-wire-verify-123` (the key came from the environment variable, not from the file) · `model: deepseek-v4-pro` · `reasoning_effort: "max"` · `stream: true` · body keys `[messages, model, reasoning_effort, stream, temperature, tools]`. - Then `{"type":"set","key":"model","value":"other-model"}`, where that model belongs to a *different* provider/group. The next request went to **that provider's** `base_url` with `Authorization: Bearer inline-other-key` and `model: other-model` — so the guide's claim that switching a grouped model moves the base URL and the key in one step is what actually happens on the wire. Scope, stated plainly: this verifies Ember's config fields and request construction. It was **not** run against `api.deepseek.com` — no DeepSeek key is available on the machine I verified from — so the DeepSeek-side behaviour in the guide comes from your own docs (`reasoning_effort` = `low|high|max`, V4 1M context) rather than from my own API calls. Happy to add a live-API confirmation as a follow-up commit. **6. No workarounds for already-fixed upstream bugs** — none suggested. ## Install paths referenced The pre-built packages link to the `v0.10.1` release assets (both verified served; the SHA256 of each is listed in the release notes), and the same bytes are also committed in-tree in the repo. Source builds cover `build.ps1` (plain g++), the CMake path (`cmake -B build -G Ninja && cmake --build build && ctest` — 23/23 passing), and `build-linux.sh`, which is what produces the released Linux binary. --- <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 09:49:23 +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#438

No description provided.

Rows
Columns