DEV Community

Deduped
Deduped

Posted on Originally published at deduped.beehiiv.com

5 hot GitHub repos this week and exactly how to use them (Nov 2, 2026)

#github #ai #opensource #productivity

Every Monday Deduped picks the five hottest new AI repos on GitHub and writes down exactly how to use each one. Here is this week's list.

1. deepseek-ai/deepseek-harness: Start the DeepSeek Harness Web UI with one command

DeepSeek Harness (dsh) is an open-source agent harness built on an everything-is-a-plugin architecture using Cordis, with a local Web UI.

Time: 15 minutes | Level: Beginner

  1. Install Node.js.
  2. Read SAFETY.md in the repository and confirm you accept the safety notice.
  3. Run npx @deepseek-ai/dsh web

Watch out: The project is in developer preview and the README warns that compatibility-breaking changes will happen.

Full steps and agent prompt

2. storytold/photocraft: Open, edit and save PSD files without Photoshop

PhotoCraft is an open-source, clean-room photo editor in pure Rust modeled on Photoshop: layers, masks, adjustment layers, layer styles, type, vectors, brushes and real PSD files, offline.

Time: 10 minutes | Level: Beginner

  1. Download PhotoCraft from its page on getartcraft.com, or build it from source with Rust (cargo run --release -p photocraft)
  2. Open a PSD; menus, panels and shortcuts follow Photoshop
  3. Edit layers, masks, adjustment layers and type
  4. Save as PSD or export to PNG, JPEG, TIFF or WebP

Watch out: Early alpha: re-saved files are not byte-identical, and some tools and generative features are missing.

Full steps and agent prompt

3. deepseek-ai/DeepInfer: Run offline inference on DeepSeek-V4.1-Flash locally

DeepInfer is an open-source, high-performance offline inference framework from DeepSeek for running DeepSeek models, currently DeepSeek-V4.1-Flash, on NVIDIA SM100 GPUs or Huawei Ascend 950 NPUs.

Time: 2 hours | Level: Advanced

  1. Create and activate an empty Python 3.12 environment.
  2. Run export DEEPINFER_ACCELERATOR=nvidia (use ascend instead on the Ascend platform), and keep it exported in the shell where you run DeepInfer.
  3. Run ./build.sh from the repo root.
  4. (Optional) Run DEEPINFER_HF_MODEL_DIR=/path/to/DeepSeek-V4.1-Flash-1017 to use a local checkpoint instead of the Hugging Face cache.
  5. Run python examples/offline_inference.py and read the output.

Watch out: The README says the open-source release covers offline inference only, and KV cache, prefill and decode disaggregation, and data-parallel load balancing are not included.

Full steps and agent prompt

4. Niko1221/Strata: Point Claude Code or Codex at a local model with strata connect

Strata runs large open models such as Qwen3.8-Flash-Next, Qwen3.6-35B-A3B and DeepSeekV4.1-Flash on your own computer, as a desktop app or a terminal app, with no account.

Time: 30 minutes | Level: Intermediate

  1. Install Strata and start a model with strata launch
  2. Run strata connect to see the setup instructions for your tool
  3. Apply the settings for Claude Code, Codex CLI, OpenCode or another listed tool
  4. Pick the Coder pack from the Models docs if you mostly write code
  5. Run a small coding task and compare speed and quality with your cloud model

Watch out: Local models are slower than cloud APIs on most consumer GPUs; the README lists 4 to 25 tokens per second for DeepSeekV4.1 on single cards.

Full steps and agent prompt

5. MemPalace/mempalace: Give Claude Code searchable memory of your past sessions

MemPalace is a local-first AI memory system.

Time: 20 minutes | Level: Beginner

  1. Run uv tool install mempalace to put the mempalace CLI in an isolated environment on your PATH (or pipx install mempalace)
  2. Run mempalace init ~/projects/myapp with the path to your own project
  3. Run mempalace mine ~/projects/myapp to mine the project files into the palace
  4. Run mempalace mine ~/.claude/projects/ --mode convos to mine your Claude Code sessions (scope with --wing per project)
  5. Run mempalace search "why did we switch to GraphQL" with your own question to find the original conversation
  6. Run mempalace wake-up to load context for a new session
  7. Run mempalace audit to score how well organized the palace is (read-only)

Watch out: The palace holds your full transcripts verbatim, so keep it on a disk you trust. The README points to its Claude Code retention setup checklist for wiring the auto-save hooks and backing up existing JSONL transcripts.

Full steps and agent prompt

Want this every week? Get the Monday email free: https://dedupedai.com/join/?utm_source=devto&utm_medium=article&utm_campaign=hot5_2026-11-02

Top comments (0)