Skip to content

PowerContext v1.0.0

Choose a tag to compare

View all tags
@Teingi Teingi released this 11 Sep 08:30
· 196 commits to master since this release
powercontext-v1.0.0
ec97efb
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

PowerContext v1.0.0

English | 中文

English

PowerContext v1.0.0 is here.

PowerContext gives humans and agents a shared place to preserve, hand off, and reuse work context. Memory keeps project decisions and constraints available across conversations. A committed Handoff carries the current objective, verified progress, evidence, and next steps, so another session or collaborator can pick up where the work left off.

v1.0.0 builds on this foundation with evolving Topic Memory, Scope profiles, and a reviewed Memory → Experience → Skill workflow. Source material, lasting knowledge, work handoffs, and reusable methods stay connected through evidence references and version history, with access governed by their respective Scopes.

Highlights

The following highlights cover changes since v0.2.0.

  • Guided setup. An interactive wizard walks through storage, deployment, memory capabilities, the Dashboard, model services, and Agent connections, then writes an environment file and tailored next steps. Basic memory supports explicit saving and recall without an additional model API. Configure the relevant models and background processing when you need automatic extraction or semantic search.

  • Topic Memory. Organize incoming Sources into durable topics with a title, summary, and detailed content. Topics evolve with new evidence while retaining provenance and revision history. Full-text, vector, and hybrid search let you retrieve a summary first and follow an exact reference to the details. Automatic processing requires the relevant model and scheduling configuration; each input does not necessarily create a new topic.

  • Scope profiles. Generate a profile from evidence within a Scope, with automatic activation or review before activation. Profiles support manual edits, revision history, and restoration. Applications can write the same subject material to a business Scope and its associated user Scope, building separate evidence histories. Profile generation supports custom prompts per Scope, and profiles can be explicitly included in Agent context.

  • Dream: turn memory into experience, then into a Skill. Select existing Memory entries to refine an Experience, or derive a Skill from approved Experience revisions. Dream preserves exact evidence references and run status, deduplicates repeated requests, and executes background work within a defined budget. It can report that no change is needed or that more evidence is required. Proposed content enters candidate review and becomes an approved revision only after approval. Exporting or installing a Skill remains an explicit action.

  • Configurable context. PreparedContext can select Memory, Experience, Profile, and Topic Memory, with control over section order, entry limits, and the total byte budget. Prompts managed per Scope let you customize instructions and examples for supported operations while retaining version history. Unified tags help organize and query Memory entries and supported Artifacts. Custom prompts preserve the operation's permission boundaries and output contract.

  • A new personal Dashboard. Read saved memories, experiences, skills, handoffs, and usage in the browser, with new Topic Memory browsing, search, and exact revision views. The Dashboard is disabled by default and uses static Bearer Token authentication when enabled. It is intended for personal use and demos; candidate review and content management use the CLI, Client, or API.

  • Agent integration improvements. A new MiniMax Code plugin provides memory and handoff workflows through MCP and a project-context Skill. Setup and diagnostics improve across Codex, Claude Code, DeepSeek Harness, and other integrations. Supported hosts can persist authorization bound to a Server URL, and Claude Code now sends the required MCP authorization header. Automatic capture, context injection, and handoff support vary by host; consult the relevant integration guide.

  • Background processing, APIs, and documentation. A unified Artifact processing supervisor manages work admission, concurrency, timeouts, and recovery from persisted state. Expanded Scope filtering and Source browsing APIs make stored context easier to discover, while Artifact generation uses consistent evidence access boundaries. Reorganized Chinese and English documentation adds static search, deployment and tracing troubleshooting, and 22 Jupyter tutorials covering Memory, handoffs, review, and team reuse. The first seven require no model API.

Fixes and reliability improvements

  • powercontext server run can load the current directory's .env, use an explicit environment file, or disable file loading, with defined configuration precedence.
  • Fixed write checks after subject Scope permissions are revoked and strengthened validation of raw paths in Skill ZIP packages.
  • Fixed Artifact conditional reads for weak ETags and multiple If-None-Match values.
  • Improved Topic Memory runtime profile checks and index reuse, DSH automatic-path diagnostics, and personal service startup on macOS.
  • Added an explicit --allow-insecure-http option for plain HTTP connections to non-loopback addresses, with address and consent persistence in supported integrations.

Install and get started

Python 3.11+, Git, and uv are required. These commands use Bash on macOS or Linux.

Install or upgrade PowerContext:

uv tool install --force "powercontext[cli,server]==1.0.0"

For a new installation, create the configuration in a separate directory and start the Server:

mkdir -p ~/powercontext-config
cd ~/powercontext-config
powercontext config init --language en --output .env
powercontext config validate --env-file .env
powercontext server run --env-file .env

Keep the Server running. In another terminal, follow the generated .env.next-steps.md to configure the client connection, create or bind a Scope, and install the matching Agent integration. For Codex, after configuring the connection and Scope, run:

powercontext setup codex --ref powercontext-v1.0.0
powercontext doctor codex

Open a new Agent session and verify saving and recall. If automatic memory is enabled, also verify Source ingestion, Topic Memory updates, and recall across sessions. See the English quick start for the complete workflow.

Upgrade and compatibility notes

  • Check migration requirements for existing databases. Back up your database and configuration before upgrading. An older database with existing Sources, Artifacts, or processing data and no completed migration marker is rejected at normal startup. This requirement is not limited to deployments that previously enabled Topic Memory. Inspect the plan with the command below. If migration is required, follow the processing migration guide: stop old processing services, pause writes, apply the migration, and verify it. Resume service after verification returns ready: true. New, empty databases initialize automatically.
  • Dashboard behavior has changed. The previous bundled UI has been replaced by a personal content viewer that is disabled by default. Enabling it requires static Bearer authentication. Multi-user deployments with custom authentication or authorization providers should use the API, MCP, or Agent integrations.
  • Upgrade connected components together. Use matching Server, Client, and Agent integration versions. Preserve and validate existing configuration, and retain any optional extras, such as seekdb, in the installation command. If you use a native personal service, run powercontext service install again.
  • Model and platform requirements still apply. Explicit Memory operations require no model API. Automatic extraction, profiles, and Dream use Server-side model configuration; vector search requires Embedding. An Agent subscription does not supply the Server's model API credentials. Windows support remains experimental, and embedded seekdb is available only on compatible macOS and Linux environments.

Read-only migration inspection for an existing deployment:

powercontext server processing-migrate --action plan --env-file /path/to/powercontext.env

Resources

中文

PowerContext v1.0.0 正式发布。

PowerContext 为人和 Agent 提供可以持续积累、交接和复用的工作上下文。项目决策和约束可以保存为 Memory，当前目标、已验证进展、证据和下一步可以提交为 Handoff。换一个会话，或把工作交给另一位协作者时，接手者能够找回相关信息，从已有进展继续。

v1.0.0 在这些能力之上，加入持续演进的 Topic Memory、Scope 画像，以及经过审核的 Memory → Experience → Skill 工作流。原始材料、长期知识、任务交接和可复用方法通过来源引用与版本记录连接起来，并受各自 Scope 的访问权限约束。

主要更新

以下为相较 v0.2.0 的主要变化。

  • 交互式配置向导。 一步步配置存储、部署场景、记忆能力、Dashboard、模型服务和 Agent 连接，生成环境文件及对应的后续操作说明。基础记忆模式无需额外模型 API，即可显式保存和召回；需要自动提取和语义检索时，再配置对应模型与后台处理。

  • Topic Memory：按主题积累和更新知识。 将持续进入的 Source 材料组织为包含标题、摘要和详情的主题记忆，随新证据更新，并保留来源与历史版本。支持全文、向量及混合检索；召回时可以先读取摘要，再按精确引用查看详情。自动处理需要启用相应模型与调度，不保证每条输入都会产生新主题。

  • Scope 画像。 根据一个 Scope 内的证据生成画像，支持自动生效或审核后生效，以及人工修改、历史版本查看和恢复。应用可以将同一份主体材料写入业务 Scope 与对应的用户 Scope，分别积累证据。画像生成支持按 Scope 自定义提示词，也可以显式加入 Agent 的上下文输出。

  • Dream：从记忆提炼经验，再形成 Skill。 选择已有 Memory 条目提炼 Experience，或基于已批准的 Experience 派生 Skill。Dream 保留精确证据引用和运行状态，支持重复请求去重及受预算约束的后台执行，也可返回无需变更或需要补充证据。生成结果进入候选审核流程，批准后才成为正式版本；Skill 的导出和安装仍需显式执行。

  • 按任务组织上下文。 PreparedContext 支持选择 Memory、Experience、Profile 和 Topic Memory，调整类别顺序、各类条数和总字节预算。新增按 Scope 管理的 Prompt，可自定义受支持操作的指令与示例并保留历史版本；统一标签用于整理和查询 Memory 条目及受支持的 Artifact。Prompt 自定义不改变工具权限或输出契约。

  • 新的个人 Dashboard。 通过浏览器阅读已保存的记忆、经验、技能、交接和用量，新增 Topic Memory 浏览、搜索与精确版本查看。Dashboard 默认关闭，启用后使用静态 Bearer Token 登录，适合个人使用和演示；候选审核和内容管理通过 CLI、Client 或 API 完成。

  • Agent 接入改进。 新增 MiniMax Code 插件，通过 MCP 和项目上下文 Skill 使用记忆与交接能力。完善 Codex、Claude Code、DeepSeek Harness 等集成的配置和诊断，为支持的宿主保存与 Server 地址绑定的授权配置，并修复 Claude Code MCP 授权头传递问题。各宿主的自动采集、上下文注入和交接能力以对应集成说明为准。

  • 后台处理、API 与文档。 引入统一的 Artifact 后台处理监督机制，管理任务准入、并发、超时及持久化恢复；扩展 Scope 筛选和 Source 浏览接口，并统一制品生成时的证据读取边界。中英文文档按用户使用流程重新组织，新增静态文档搜索、部署与追踪排障说明，以及 22 篇 Jupyter 教程，覆盖 Memory、交接、审核和团队复用；前七篇无需模型 API。

修复与可靠性改进

  • powercontext server run 支持自动加载当前目录的 .env、显式指定环境文件或关闭文件加载，并明确配置优先级。
  • 修复主体 Scope 权限撤销后的写入检查，并加强 Skill ZIP 包原始路径校验。
  • 修复 Artifact 条件读取对弱 ETag 和多值 If-None-Match 的处理。
  • 改进 Topic Memory 运行时配置检查与索引复用、DSH 自动路径诊断，以及 macOS 个人服务启动行为。
  • 为非环回地址的明文 HTTP 连接提供显式 --allow-insecure-http 选择，并在支持的集成中保存地址与同意状态。

安装与开始使用

需要 Python 3.11+、Git 和 uv。以下命令使用 Bash，适用于 macOS 和 Linux。

安装或升级 PowerContext：

uv tool install --force "powercontext[cli,server]==1.0.0"

首次安装可在独立目录中生成配置并启动 Server：

mkdir -p ~/powercontext-config
cd ~/powercontext-config
powercontext config init --language zh --output .env
powercontext config validate --env-file .env
powercontext server run --env-file .env

保持 Server 运行，在另一个终端按照生成的 .env.next-steps.md 配置客户端连接、创建或绑定 Scope，并安装匹配版本的 Agent 集成。以 Codex 为例，完成连接与 Scope 配置后执行：

powercontext setup codex --ref powercontext-v1.0.0
powercontext doctor codex

打开新的 Agent 会话，验证保存与召回；启用自动记忆后，还应检查 Source 入库、Topic Memory 更新及跨会话召回。完整步骤见中文快速开始。

升级与兼容性说明

  • 已有数据库需要检查迁移。 升级前备份数据库和配置。旧数据库中已有 Source、Artifact 或后台处理数据且缺少迁移完成标记时，普通启动会拒绝运行；这一要求不限于曾启用 Topic Memory 的部署。先使用下方命令查看计划。需要迁移时，按后台处理迁移指南停止旧处理进程、暂停写入，再执行迁移和验证；验证返回 ready: true 后恢复服务。全新的空数据库会自动初始化。
  • Dashboard 行为发生变化。 原内置 UI 已替换为默认关闭的个人内容查看器。启用它需要静态 Bearer 鉴权；使用自定义认证或授权 Provider 的多用户部署应通过 API、MCP 或 Agent 集成访问内容。
  • 同步升级连接组件。 Server、Client 和 Agent 集成应使用匹配版本。已有部署保留并校验原配置；启用了 seekdb 等可选依赖时，安装命令中也需保留对应 extra。使用原生个人服务的用户应重新执行 powercontext service install。
  • 模型与平台要求保持明确。 显式 Memory 操作无需模型 API；自动提取、画像和 Dream 等生成能力使用 Server 侧模型配置，向量检索需要 Embedding。Agent 订阅不会自动提供 Server 的模型 API 凭据。Windows 支持仍为实验性，嵌入式 seekdb 仅支持兼容的 macOS 和 Linux 环境。

已有部署的只读迁移检查：

powercontext server processing-migrate --action plan --env-file /path/to/powercontext.env

相关资源

Assets 4
Loading