DeepSeek open-sources deepseek-recipe, its production token encoder/decoder matching official DeepSeek API behavior
DeepSeek released deepseek-recipe as open source, a production token encoding and decoding component. It replicates the processing logic behind the official DeepSeek API so community deployments can match its API behavior.
- DeepSeek releases deepseek-recipe, a token encoding/decoding component aligned with official DeepSeek API processing logic for community deployments
- deepseek-recipe brings the same API behavior as DeepSeek's official service to self-hosted community deployments
Entities
Related
Any coding agent vendors actually using DeepInfer or deepseek-recipe yet?
A developer asks whether any coding agent vendors publicly use DeepSeek's open-sourced DeepInfer inference engine or deepseek-recipe, seeking self-hosting that matches the official API for DeepSeek V4.1 Flash 1017 without manual CUDA work. No vendor adoption is reported in the thread.
A developer asks whether any coding agent vendors publicly use DeepSeek's open-sourced DeepInfer inference engine or deepseek-recipe, seeking self-hosting that matches the official API for DeepSeek V4.1 Flash 1017 without manual CUDA work. No vendor adoption is reported in the thread.
DeepSeek releases Harness v0.1 agent harness in developer preview, MIT-licensed
DeepSeek announced Harness v0.1, an agent harness now in developer preview and open-sourced under the MIT license. Built on the Cordis meta-framework, it targets developers building agent harnesses worldwide. The release marks DeepSeek's entry into the agent tooling space.
DeepSeek announced Harness v0.1, an agent harness now in developer preview and open-sourced under the MIT license. Built on the Cordis meta-framework, it targets developers building agent harnesses worldwide. The release marks DeepSeek's entry into the agent tooling space.
CLI that cuts AI coding token usage by up to 97%
A command-line tool that reduces token consumption when using LLMs for coding tasks. It targets developers using AI coding assistants who face high token costs. The post claims token usage reductions of up to 97%.
A command-line tool that reduces token consumption when using LLMs for coding tasks. It targets developers using AI coding assistants who face high token costs. The post claims token usage reductions of up to 97%.
Fulmar: native macOS app wrapping DeepSeek Harness agent runtime with access, model-routing and recovery controls
Fulmar is a native Mac app that wraps the DeepSeek Harness (DSH) agent runtime, which handles conversations, tools and tasks. It adds management controls the raw runtime lacks: choosing local or cloud model routes with explicit consent, Keychain credential storage, workspace/tool access limits, and recovery checkpoints. It targets developers running file- and command-capable agents on macOS who want oversight and safety.
Fulmar is a native Mac app that wraps the DeepSeek Harness (DSH) agent runtime, which handles conversations, tools and tasks. It adds management controls the raw runtime lacks: choosing local or cloud model routes with explicit consent, Keychain credential storage, workspace/tool access limits, and recovery checkpoints. It targets developers running file- and command-capable agents on macOS who want oversight and safety.
Anthropic open-sources distillation check tool
Anthropic released an open-source tool for checking distillation quality in AI models. The tool helps practitioners verify that distilled models retain fidelity to their teachers.
Anthropic released an open-source tool for checking distillation quality in AI models. The tool helps practitioners verify that distilled models retain fidelity to their teachers.

