Release PowerContext v1.2.0 · oceanbase/powercontext

Not only memory, but a full story.

Release tag: powercontext-v1.2.0 · Python package: powercontext==1.2.0

English

PowerContext v1.2.0 helps coding agents connect the code in front of them with the project knowledge accumulated over time. This release introduces native code intelligence, a lightweight Jev integration, Memory quality and capacity controls, and broader Agent integrations.

1. Code Intelligence for Coding Agents

Give your Agent a clearer view of the repository it is changing.

Index mixed Python, JavaScript, TypeScript, and Go repositories. Find definitions, inspect static relationships and potentially affected tests, and retrieve source evidence through CLI, HTTP, Python Client, Runtime, and MCP.
Bring code evidence and project history into the same bounded context, with optional injection through Codex and Claude Code hooks.
Refresh the index as code changes. Queries detect changed source content and reject stale evidence, helping Agents work from the current implementation.
Run structural analysis without a generation or embedding model. Embedded seekdb can store the code graph; other deployments use a separate SQLite cache.

Code intelligence requires the code extra and explicit configuration. Local indexing has been verified on Linux; macOS and Windows local indexing remain unverified. Static relationships identify leads for investigation; actual behavior still needs testing.

Explore native code intelligence

2. Plug Jev into the Decision Loop

Let recalled project rules and experience inform model decisions.

The new Jev/Laya example connects Jev through a lightweight adapter and PowerContext's DecisionModel interface. The generation model is configured independently, so applications can add Jev's judgments to their existing coding workflow without changing PowerContext's core APIs or storage schema.

The runnable example demonstrates a complete loop:

Save project rules → recall them in a fresh process → generate code with and without context → ask Jev to assess both implementations → run independent tests → save and recall the observed outcome.

Runtime-managed timeouts and fallback handling make provider failures explicit. Optional Laya support uses the same decision interface. Jev's judgments remain advisory; the tests establish what the generated code actually does.

The adapter, configuration templates, and runnable example are available in examples/systemone.

3. Memory Quality & Capacity Control

Make long-lived Memory easier to trust and manage.

Check evidence before writing. An optional Memory write gate assesses proposed content against its cited evidence, can hold unsupported writes, and exposes the reason through operation results and diagnostics. Decision assistance and the write gate are disabled by default; backend failure or abstention falls back to the existing write path.
Inspect and bound growth. Capacity reporting exposes active entries, manifest entries, and content size. Default budgets per Memory Revision are 5,000 active entries, 10,000 manifest entries, and 4 MiB of canonical content. Writes that further increase an exceeded dimension return memory_capacity_exceeded.
Compact the current manifest while retaining history. Optional compaction removes eligible, untagged inactive entries from the current manifest. It is disabled by default, preserves historical revisions, and does not impose a total database disk limit.

4. Broader Agent Integration

Bring project context into more of your everyday tools.

ZCode: A new community, experimental integration connects the CLI and Windows desktop app, with setup and diagnostics, prompt capture, context injection, MCP operations, and Handoff workflows.
Codex: Generate Experience and Skill candidates, discover and import external Skills for review, and bind a checkout to an existing Scope for later sessions. Native MCP can reuse saved credentials, and diagnostics distinguish basic connectivity from complete tool coverage. Candidate approval remains explicit; importing or approving a Skill does not install or execute it.
Hermes, WorkBuddy, and DeepSeek Harness: Improve pre-compression checkpoints, recall of the submitted turn, and cold-start readiness handling.

Agent integration guides

More improvements

Incremental Memory search-index updates reuse unchanged embeddings.
Memory errors retain their structured details over HTTP; Topic Memory publication works on SQLite builds without RETURNING support.
More flexible embedding-provider compatibility and bounded Qwen thinking configuration for Topic Memory.
Model usage accounting runs outside inference deadlines, so accounting failures do not fail a successful model operation.
Paired continuation evaluation with Codex and Bub, plus LoCoMo-Plus evaluation and a LongMemEval-V2 smoke adapter.
More reliable native service startup and automated release-version maintenance.

Install or upgrade

Install the CLI and Server:

uv tool install --force "powercontext[cli,server]==1.2.0"

To include native code intelligence, use powercontext[cli,server,code]==1.2.0 instead and follow the code configuration guide above. Preserve any other extras your deployment uses, such as seekdb.

For a new installation, follow the Quick Start to configure the Server connection and Scope. Install the matching Agent integration, for example with Codex:

powercontext setup codex --ref powercontext-v1.2.0
powercontext doctor codex

For existing deployments:

Back up the database and configuration, upgrade the Server, clients, and Agent integrations together, restart the Server, and open a new Agent session. Check existing Memory against the new capacity defaults before resuming growth.
Enable code indexing, decision assistance, the write gate, and compaction explicitly as needed. Calibrate the write gate before applying it to automatic extraction: held extraction windows advance their cursor and are not automatically retried.
Complete any unfinished Artifact processing migration. Windows support remains experimental; consult each integration guide for its verified coverage.

Full changelog: v1.1.0 → v1.2.0

中文

PowerContext v1.2.0，让 Coding Agent 把眼前的代码与持续积累的项目知识联系起来。本次发布带来原生代码理解、Jev 轻量接入、记忆质量与容量治理，以及更多 Agent 集成，让项目上下文更好地参与实际开发。

1. 让 Agent 更懂你的代码库

为 Coding Agent 提供更贴近当前仓库的代码上下文。

支持 Python、JavaScript、TypeScript 和 Go 混合仓库索引，通过 CLI、HTTP、Python Client、Runtime 和 MCP 查找定义、查看静态关系与可能受影响的测试，并获取源码证据。
将代码证据与项目历史一起组织进有预算限制的上下文，支持在 Codex、Claude Code Hook 中按需开启注入。
代码变化后可刷新索引；查询会检测源码变化并拒绝过期证据，帮助 Agent 依据当前实现开展工作。
结构分析无需生成模型或 Embedding 模型。嵌入式 seekdb 可直接存储代码图，其他部署使用独立 SQLite 缓存。

代码理解需要安装 code 扩展并显式配置。本地索引已在 Linux 验证，macOS 与 Windows 本地索引尚未验证。静态关系用于提供排查线索，实际行为仍需通过测试确认。

了解原生代码理解

2. Jev 丝滑接入决策回路

让召回的项目规则与历史经验参与模型判断。

新增 Jev/Laya 可运行示例，通过轻量适配器和 PowerContext 的 DecisionModel 接口接入 Jev。代码生成模型可独立配置，应用可以在已有开发流程中加入 Jev 的判断，无需改动 PowerContext 核心 API 或存储结构。

示例展示完整的工作闭环：

保存项目规则 → 在新进程中召回 → 对比有无上下文的代码生成 → Jev 判断两份实现 → 独立测试验证 → 写回并再次召回实际结果。

Runtime 统一处理超时与故障降级，并明确标记降级结果；同一决策接口也支持可选的 Laya 接入。Jev 提供判断建议，独立测试验证生成代码的实际行为。

适配器、配置模板与运行说明见 examples/systemone。

3. 记忆质量门禁与容量治理

让长期记忆的写入有据可查，增长有界可管。

写入前检查证据。 新增可选的 Memory 写入门禁，在保存前检查候选内容与引用证据的支撑关系，可拦截证据不足的写入，并通过操作结果和诊断信息说明原因。Decision assistance 与写入门禁均默认关闭；判断模型故障或弃权时，回退原有写入流程。
看得见、管得住容量。 可查看活跃条目数、清单条目数和内容大小。每个 Memory Revision 默认预算为 5,000 条活跃条目、10,000 条清单条目和 4 MiB 规范内容；继续扩大超限维度的写入返回 memory_capacity_exceeded。
整理当前清单，保留历史记录。 可选压缩会移除当前清单中符合条件且未加标签的失活条目。该功能默认关闭，保留历史 Revision，不限制整个数据库的磁盘占用。

4. Agent 接入持续扩展

让项目上下文进入更多日常开发工具。

ZCode： 新增社区维护的试验性集成，覆盖 CLI 与 Windows 桌面端，支持安装诊断、提示采集、上下文注入、MCP 操作和 Handoff 工作流。
Codex： 支持生成 Experience、Skill 候选，发现并导入外部 Skill 供审核；可将工作区绑定到已有 Scope，供后续会话继续使用。原生 MCP 可复用已保存的凭据，诊断也会区分基础连接与完整工具覆盖。候选仍需显式审核，导入或批准 Skill 不会自动安装或执行。
Hermes、WorkBuddy 与 DeepSeek Harness： 完善压缩前检查点、本轮提示召回与冷启动就绪探测。

查看 Agent 接入指南

安装与升级

安装 CLI 与 Server：

uv tool install --force "powercontext[cli,server]==1.2.0"

如需原生代码理解，将安装项替换为 powercontext[cli,server,code]==1.2.0，并按上方代码理解指南完成配置。保留部署已使用的其他扩展，例如 seekdb。

首次安装请按快速开始配置 Server 连接与 Scope，再安装对应版本的 Agent 集成。以 Codex 为例：

powercontext setup codex --ref powercontext-v1.2.0
powercontext doctor codex

已有部署升级时：

备份数据库与配置，一起升级 Server、客户端和 Agent 集成，重启 Server 并打开新的 Agent 会话。继续增加记忆前，先检查现有内容是否超过新增的默认容量预算。
按需显式开启代码索引、Decision assistance、写入门禁和压缩。将写入门禁用于自动提取前，应先完成校准：被拦截的提取窗口仍会推进游标，不会自动重试。
完成尚未结束的 Artifact 处理迁移。Windows 支持仍为试验性，各 Agent 的已验证范围请查看对应集成指南。

完整变更：v1.1.0 → v1.2.0

What's Changed

fix(windows): make worker pipe typing portable by @Doar999 in #1682
build(deps): bump docker/setup-qemu-action from 4.3.0 to 4.4.0 by @dependabot[bot] in #1686
build(deps): bump docker/setup-buildx-action from 4.3.0 to 4.4.1 by @dependabot[bot] in #1687
fix(service): avoid background throttling on macOS by @sunlight798 in #1694
docs(topic-memory): align the guide with the shipped generic Artifact APIs by @AlexStocks in #1690
docs(get-started): add FAQ covering Memory, Handoff, Experience, and … by @guamiao in #1697
fix(cli): align server port and agent setup configuration by @longdafeng in #1696
feat(benchmark): add LoCoMo-Plus evaluation by @sunlight798 in #1693
docs(config): document the recall-gate Runtime keys by @AlexStocks in #1689
docs: add memory quality,forgetting and lifecycle RFC by @jiannnnyyyyy in #1652
fix(service): bound the install readiness wait by the native job lifecycle by @AlexStocks in #1704
fix(recall): honor the query contract bound and the remote embedding budget by @AlexStocks in #1700
fix(config): suggest port 17429 in the setup wizard by @PsiACE in #1681
fix(codex): reuse saved credentials for native MCP by @Teingi in #1706
fix(server): preserve canonical memory error details by @DongYao0 in #1712
docs(get-started): extend FAQ with Middleware and MCP tools by @guamiao in #1710
perf(memory): update search projections incrementally per revision by @petercheng in #1709
fix(evaluation): register a Scope for each arm before running Codex by @Fengzdadi in #1723
docs(get-started): add FAQ for importing and forking external Skills by @guamiao in #1726
fix(uv): stop the integrations from constraining powercontext's version by @jiannnnyyyyy in #1728
docs: clarify implementation and code review guidelines by @PsiACE in #1721
fix(workbuddy): keep prompt hooks from stalling on scope reads by @AlexStocks in #1729
feat(code): add native repository understanding by @Teingi in #1708
fix(workbuddy): reduce the host prompt to the submitted turn before recall by @AlexStocks in #1703
chore(github): add CODEOWNERS by @zhanghuidinah in #1735
docs(integrations): update timeout defaults by @AlexStocks in #1741
fix(hermes): skip automatic writes outside primary agent contexts by @AlexStocks in #1738
fix(dsh): allow cold readiness probes to finish by @Inference1 in #1744
refactor(e2e): extract a host adapter from the Bub harness by @Fengzdadi in #1743
feat(hermes): honor the pre-compress checkpoint API v2 contract by @AlexStocks in #1737
feat(runtime): add decision model contracts by @AlexStocks in #1739
docs(rfc): propose a decision-model rerank seam by @AlexStocks in #1745
Codex/longmemeval v2 adapter by @chenncy in #1714
feat(runtime): add a decision model seam with opt-in and fail-open by @AlexStocks in #1740
fix(inference): let embedding requests omit dimensions by @knqiufan in #1755
build(deps): bump astral-sh/setup-uv from 10.0.0 to 10.2.0 by @dependabot[bot] in #1758
fix(sqlite): preserve old SQLite compatibility for Topic Memory by @knqiufan in #1764
feat(memory): enforce capacity budgets for long-lived Memories by @222twotwotwo in #1746
feat: Enhance Codex and related integrations with Experience and Skill capabilities by @Alanxtl in #1765
docs(rfc): define usability and agent workflows by @PsiACE in #1733
feat(codex): bind a checkout to an existing Scope by @knqiufan in #1760
fix(topic-memory): allow bounded Qwen thinking override by @knqiufan in #1763
fix(usage): keep model usage accounting out of model deadlines by @guamiao in #1734
feat(zcode): integrate PowerContext with ZCode CLI and Windows desktop by @jackie-cqz in #1752
feat(e2e): compare PowerContext off and on for Bub continuation workloads by @Fengzdadi in #1747
fix(examples): make Experience content assertions robust to server-added optional fields by @xiaodu55 in #1775
feat: update plugin version to 1.1.0 and refactor user-agent headers to use dynamic versioning by @Alanxtl in #1773
feat(e2e): run paired continuation workloads with Codex by @Fengzdadi in #1779
feat(memory): gate writes on decision-model evidence sufficiency by @AlexStocks in #1742
chore(release): prepare v1.2.0 and automate version updates by @Teingi in #1777
feat(examples): Jev/Laya into PowerContext's decision workflow by @Teingi in #1797

New Contributors

@guamiao made their first contribution in #1697
@DongYao0 made their first contribution in #1712
@petercheng made their first contribution in #1709
@Fengzdadi made their first contribution in #1723
@Inference1 made their first contribution in #1744
@jackie-cqz made their first contribution in #1752
@xiaodu55 made their first contribution in #1775

Full Changelog: powercontext-v1.1.0...powercontext-v1.2.0

Navigation Menu

PowerContext v1.2.0

Choose a tag to compare

Sorry, something went wrong.