有问题,先来这里找答案

开发者文档与帮助中心一站两区:接入 tansr 的每一步,从安装到排障,都在这里。

浏览全部分类前往帮助中心

按分类浏览

Popular articles

Technical stance

Collapsed by default. The roadmap is separate from what is already offered.Technical long form Plain-text copy

Bridge: Tansr is the product direction used to connect models, data, tools, business logic and the runtime. It is a bridge between agents and people: existing model capability is organized into product capability that can be integrated, reused and kept in development. Tansr does not turn itself into another model vendor.
Neutrality: Neutrality is an integration boundary. The runtime is not tied to one model vendor, and the product is not defined by replacing a vendor. Vendor compute and models are connected to real demand; the developer decides business data, interface, sign-in and authorization. A model named in the docs or an example shows one integration. It does not exclude others, and it does not mean every model, platform and industry is already supported.
Harness: The Harness is the runtime in which an agent understands a task, arranges steps, calls tools and keeps going. The CLI, the Node SDK, the session service, and the Android, iOS and HarmonyOS clients are different forms of one kernel: the query loop, tool scheduling, the permission engine and context compaction. Placing a logo on a screen, or calling a model once while bypassing that kernel, is not a Harness integration.
Division of work: Tansr Skill is installed into the coding assistant the user already uses. It helps that assistant understand the request, plan the product, connect capabilities, build the application and check the result. It is not a runtime skill inside the application, and it is not the Node SDK. At runtime the SDK and the session service carry task execution, tool calls, sessions, context, permissions and usage. The interface, business data, user sign-in, business authorization, and any price charged to an end user, stay in the application. Installing the Skill does not create a platform account, call a model, or deploy an application.
The current two steps: The current path has two steps. First install the published Tansr Skill, either with npx or by giving the install instruction to the same assistant. Then describe the product to that assistant. The product can be new, or AI added to one that already exists. The two steps exist to cut the concepts required at the start, so a person can state the need first. They are not two clicks that publish a product, and they do not mean there is no account, no configuration, unlimited calls, or a fixed duration.
What counts as a product: The deliverable is a product that can be operated, checked, changed and developed further. It needs an interaction or call entry that fits the target form, a business flow, and a visible result of tool execution. Forms include a graphical interface, the CLI, a Node process and the session service. A product is not limited to one kind of interface. A chat reply, a static preview, or an SDK fragment alone is not a finished product.
Four depths: The same capabilities cover different starting points. A person building a product for the first time gets continuous guidance from one request. A person who already writes code reuses model calls, tool execution, sessions and permissions instead of rebuilding the agent substrate. A team that keeps architectural control integrates along its own business logic, data and interface. An organization cares about applications, permissions, usage and continued operation. These four depths are ways in. They are not four plans, and nobody has to pick an identity first.
Four parties: The connection gives each party a technical place. Model vendors get an outlet into real products. Developers reuse a general agent runtime instead of implementing the query loop and the permission engine from nothing. End users meet intelligence through a concrete product, not by learning the protocol first. Agents get an environment in which they can act, and a way into different application forms. A platform, industry or model that a page does not name is not thereby supported.
Roadmap and what exists: The long-term path has four steps. Anything not built yet is a roadmap, not a shipped capability. Step one is the work now: use the CLI, the SDK and the session service to make the Harness reliable enough to act. Step two is SDKs and services on more platforms, and opening the core. The MIT license of the Skill, and the MIT license on the SDK and session service packages, do not mean the whole kernel is already published under that license. The CLI package keeps its own license. Step three, agents collaborating on a platform and gradually taking part in its governance and operation, is not a product today. Step four, people and agents carrying that capability into daily life and industry, is not a feature today. A Tansr-owned coding assistant, and an entry that only asks for a product description and a target platform, are not offered.
Published packages only: The docs site describes behavior you can install. The published Tansr Skill release is 0.1.6. Versions of the CLI, the SDK, the session service and the other language clients are the ones named on their own pages and in those packages' release notes. Changes that exist only on the main branch, including higher candidate version numbers, are not capabilities of the package you install today.