Teingi · Teingi · Aug 27, 2026 · Aug 27, 2026 · Aug 27, 2026 · Aug 27, 2026
diff --git a/README.md b/README.md
@@ -41,7 +41,8 @@ powercontext setup pi --source oceanbase/powercontext --ref master
 powercontext setup workbuddy --source oceanbase/powercontext --ref master
 
 # Or install several hosts in one pass.
-powercontext setup select --host codex --host claude-code --host opencode
+powercontext setup select --host codex --host claude-code --host opencode \
+  --source oceanbase/powercontext --ref master
 ```
 
 The first command installs the CLI and local Server from the latest `master` revision in an isolated environment.
@@ -60,7 +61,8 @@ In another terminal, verify the service and plugin:
 
 ```bash
 powercontext doctor
-powercontext doctor codex  # or: claude-code / dsh / hermes / openclaw / opencode / pi / workbuddy
+powercontext doctor integrations
+powercontext doctor codex  # Replace codex with the host you installed.
 ```
 
 By default, the Server listens on `127.0.0.1:8000`, exposes Streamable HTTP MCP at `/mcp`, and persists data in a

diff --git a/README_CN.md b/README_CN.md
@@ -39,7 +39,8 @@ powercontext setup pi --source oceanbase/powercontext --ref master
 powercontext setup workbuddy --source oceanbase/powercontext --ref master
 
 # 或者一次性安装多个 Host。
-powercontext setup select --host codex --host claude-code --host opencode
+powercontext setup select --host codex --host claude-code --host opencode \
+  --source oceanbase/powercontext --ref master
 ```
 
 第一条命令会在隔离环境中从最新 `master` revision 安装 CLI 和本地 Server；每条 setup 命令都会从同一个
@@ -57,7 +58,8 @@ powercontext server run
 
 ```bash
 powercontext doctor
-powercontext doctor codex  # 或: claude-code / dsh / hermes / openclaw / opencode / pi / workbuddy
+powercontext doctor integrations
+powercontext doctor codex  # 请把 codex 换成已安装的宿主。
 ```
 
 默认情况下，Server 监听 `127.0.0.1:8000`，在 `/mcp` 提供 Streamable HTTP MCP，并将数据持久化到本地

diff --git a/README_JP.md b/README_JP.md
@@ -41,7 +41,8 @@ powercontext setup pi --source oceanbase/powercontext --ref master
 powercontext setup workbuddy --source oceanbase/powercontext --ref master
 
 # 複数の Host を一度にインストールすることもできます。
-powercontext setup select --host codex --host claude-code --host opencode
+powercontext setup select --host codex --host claude-code --host opencode \
+  --source oceanbase/powercontext --ref master
 ```
 
 最初のコマンドは、隔離された環境に最新の `master` revision から CLI とローカル Server をインストールします。
@@ -60,7 +61,8 @@ powercontext server run
 
 ```bash
 powercontext doctor
-powercontext doctor codex  # または: claude-code / dsh / hermes / openclaw / opencode / pi / workbuddy
+powercontext doctor integrations
+powercontext doctor codex  # codex をインストールした Host 名に置き換えてください。
 ```
 
 デフォルトでは、Server は `127.0.0.1:8000` で待ち受け、`/mcp` で Streamable HTTP MCP を公開し、

diff --git a/docker/README.md b/docker/README.md
@@ -16,12 +16,16 @@ Run the Server with persistent SQLite and scheduler data:
 ```bash
 docker run --rm \
   --name powercontext-server \
-  --publish 8000:8000 \
+  --publish 127.0.0.1:8000:8000 \
   --volume powercontext-data:/data \
   powercontext-server:local
 ```
 
-The image listens on `0.0.0.0:8000`, stores its default data under `/data`, and exposes a Docker health check backed
+The image listens on `0.0.0.0:8000` inside the container. Keep the host-side publish address on loopback unless bearer
+authentication and a TLS-terminating network boundary are configured. See
+[`Deploy the Server`](../docs/en/docs/how-to/deploy-server.md) for the remote-access setup.
+
+The image stores its default data under `/data` and exposes a Docker health check backed
 by `GET /health/ready`. Runtime or database failures return `not_ready` with HTTP 503. A configured inference failure
 returns `degraded` with HTTP 200, so database-backed operations remain in traffic while the response exposes the
 affected capability. Provider checks make one minimal real request at startup. `ready` and `misconfigured` results
@@ -42,6 +46,6 @@ TLS-terminating proxy and enable bearer authentication with
 opt-in is no longer required.
 
 The `Build Docker image` GitHub workflow builds downloadable Linux amd64 and arm64 image archives for pull requests,
-changes merged to `main`, and manual runs. Publishing a GitHub Release pushes a multi-platform image to Docker Hub.
+changes merged to `master`, and manual runs. Publishing a GitHub Release pushes a multi-platform image to Docker Hub.
 Repository configuration must provide `DOCKER_USERNAME` and `DOCKER_PASSWORD` secrets plus a `DOCKER_PUSH_BASE`
 variable such as `oceanbase`; Release tags must use `vX.Y.Z` or `X.Y.Z` semantic versioning.
diff --git a/docs/en/docs/how-to/configure-hermes.md b/docs/en/docs/how-to/configure-hermes.md
@@ -0,0 +1,102 @@
+---
+title: Configure Hermes
+description: Install the Hermes MemoryProvider and slash-command companion, then connect them to PowerContext Server.
+---
+
+# Configure Hermes
+
+The integration contains a standard Hermes `MemoryProvider` and a standalone slash-command plugin. Hermes remains
+responsible for the conversation and memory lifecycle; the provider sends recall, capture, and explicit operations to
+a separately running PowerContext Server. The companion registers `/pc` and `/powercontext` before the provider is
+activated. Backend failures do not interrupt the Hermes conversation.
+
+## Prerequisites
+
+- Hermes Agent 0.20.4 or newer, available on `PATH`;
+- PowerContext CLI and Server installed from `master`;
+- a running PowerContext Server.
+
+```bash
+uv tool install --force "powercontext[cli,server] @ git+https://github.com/oceanbase/powercontext.git@master"
+powercontext server run
+```
+
+## Install both plugins
+
+In another terminal, install or refresh both plugins from the matching revision:
+
+```bash
+powercontext setup hermes --source oceanbase/powercontext --ref master
+powercontext doctor hermes
+```
+
+The setup command copies the provider to `$HERMES_HOME/plugins/powercontext`, installs
+`$HERMES_HOME/plugins/powercontext-command`, and enables the companion without granting built-in tool override
+permissions. It does not start the Server or select the provider in Hermes.
+
+Run the Hermes memory setup wizard and select `PowerContext`:
+
+```bash
+hermes memory setup
+```
+
+On Hermes 0.20.4, use the generic command above. `hermes memory setup powercontext` selects the provider but does not
+open its configuration wizard. Restart Hermes after setup.
+
+## Verify recall and writes
+
+```bash
+hermes powercontext status
+hermes powercontext remember preference "The user prefers uv"
+hermes powercontext search "Python package manager"
+```
+
+Inside an interactive Hermes session, `/pc status` should reach the same active provider. Use `/pc ` followed by
+Tab/Down to inspect the available Memory, Handoff, Experience, Skill, review, statistics, trace, and Workstream
+commands. Hermes 0.20.4 does not provide enough invocation context to route gateway slash commands safely, so the
+companion rejects gateway invocations; use the provider's Hermes tools in gateway sessions.
+
+The provider uses `http://127.0.0.1:8000` by default. In a Git workspace, Workstream persistence first reads the
+shared `.git/powercontext/codex-workspace.json` scope binding. An explicit scope configuration takes precedence.
+Without either value, the provider derives a scope from the active Hermes profile and gateway user identifier; for a
+local CLI session without a user identifier, it derives a stable value from `HERMES_HOME`.
+
+## Configure the connection
+
+The wizard writes non-sensitive settings to `$HERMES_HOME/powercontext/config.json`. Environment variables override
+the file:
+
+| Variable | Purpose |
+| --- | --- |
+| `POWERCONTEXT_HERMES_CONFIG` | Config file path; defaults to `$HERMES_HOME/powercontext/config.json` |
+| `POWERCONTEXT_HERMES_BASE_URL` | PowerContext Server URL |
+| `POWERCONTEXT_HERMES_AUTHORIZATION` | Complete authorization header, such as `Bearer <token>` |
+| `POWERCONTEXT_HERMES_TOKEN` | Bare-token shorthand used when `AUTHORIZATION` is absent |
+| `POWERCONTEXT_HERMES_SCOPE_ID` | Explicit scope or scope template |
+| `POWERCONTEXT_HERMES_MAX_BYTES` | Maximum prepared-context size, from 512 to 32768 bytes |
+| `POWERCONTEXT_HERMES_TIMEOUT` | HTTP request timeout in seconds |
+| `POWERCONTEXT_HERMES_CAPTURE_TURNS` | Capture completed turns as Sources |
+| `POWERCONTEXT_HERMES_FLUSH_ON_SESSION_END` | Run Memory extraction at session end |
+| `POWERCONTEXT_HERMES_CAPTURE_PRE_COMPRESS` | Capture filtered new turns before compression; disabled by default |
+| `POWERCONTEXT_HERMES_EVALUATION_TRACE` | Record recalled context in sensitive local JSONL traces; disabled by default |
+| `POWERCONTEXT_HERMES_EVALUATION_TRACE_PATH` | Override the evaluation trace directory |
+| `POWERCONTEXT_HERMES_WORKSTREAM` | Read the shared Git-private Workstream binding; enabled by default |
+
+Let the Hermes wizard store authorization in its protected `.env` secret store; do not put the token in
+`config.json`. Use plain HTTP only for a loopback Server. See [Deploy the Server](deploy-server.md) before connecting
+to a remote deployment. Evaluation traces contain prompts and recalled context; keep them local and protect them as
+sensitive data.
+
+## Enable automatic extraction only when needed
+
+Completed-turn capture creates Source evidence. It does not create Memory by itself. Automatic Source-to-Memory
+extraction requires a generation model on the Server:
+
+```bash
+export POWERCONTEXT_SERVER_INFERENCE_GENERATION_MODEL=provider:model-name
+powercontext server run
+powercontext capabilities
+```
+
+The capability output must report Memory extraction as enabled. Explicit `hermes powercontext remember` writes do not
+require a model.
diff --git a/docs/en/docs/how-to/configure-langchain.md b/docs/en/docs/how-to/configure-langchain.md
@@ -14,17 +14,18 @@ The middleware uses LangChain's public `AgentMiddleware` API. Recalled content m
 
 ## Install
 
-The middleware ships in the standalone `powercontext-langchain` distribution and requires LangChain 1.3 or later:
+The middleware source is packaged separately as `powercontext-langchain` and requires LangChain 1.3 or later. It is
+not currently published on PyPI:
 
 ```bash
-uv tool install "powercontext[cli,server]==0.0.2"
+uv tool install --force "powercontext[cli,server] @ git+https://github.com/oceanbase/powercontext.git@master"
 powercontext server run
 ```
 
 Keep the Server running, then install the middleware in the LangChain application's environment:
 
 ```bash
-uv pip install "powercontext-langchain @ git+https://github.com/oceanbase/powercontext.git#subdirectory=integrations/langchain"
+uv pip install "powercontext-langchain @ git+https://github.com/oceanbase/powercontext.git@master#subdirectory=integrations/langchain"
 ```
 
 Skip the Server installation when the application already connects to a separately managed Server. From a repository

diff --git a/docs/en/docs/how-to/configure-langgraph.md b/docs/en/docs/how-to/configure-langgraph.md
@@ -14,13 +14,12 @@ primitives that are stable public API. It never starts or embeds the Server.
 The package is not yet published to PyPI, so install it from source alongside a running Server:
 
 ```bash
-uv pip install "powercontext-langgraph @ git+https://github.com/oceanbase/powercontext.git#subdirectory=integrations/langgraph"
+uv pip install "powercontext-langgraph @ git+https://github.com/oceanbase/powercontext.git@master#subdirectory=integrations/langgraph"
 powercontext server run
 ```
 
-From a checkout you can install the local path instead: `uv pip install ./integrations/langgraph`. Publishing to
-PyPI is pending a standalone build and release step that advances the package version independently of the root
-`powercontext` distribution; until that lands, use the source install above.
+From a checkout you can install the local path instead: `uv pip install ./integrations/langgraph`. The adapter is not
+currently published on PyPI, so use one of these source installations.
 
 The package depends on `powercontext[client]`, `langgraph`, `langchain-core`, and `pydantic-settings`. It does not
 pull in the Server; point it at a Server you run separately.
@@ -148,10 +147,10 @@ Server. Implementing only search and raising for the rest produces an object tha
 fails at runtime inside unrelated nodes or tools, which is worse than providing no store. The adapter therefore
 integrates at the node and tool level and does not occupy the `store` parameter of `compile()`.
 
-## Scope of this release
+## Current scope
 
-In scope: Memory read and write, and bounded context preparation.
+Included: Memory read and write, and bounded context preparation.
 
-Out of scope for this release: automatic trajectory capture, checkpointing, Handoff, Artifact Candidate review, and
-Experience or Skill generation. Use `powercontext_remember` for explicit writes; automatic capture of a run as Source
-evidence is not part of this adapter.
+Not included: automatic trajectory capture, checkpointing, Handoff, Artifact Candidate review, and Experience or
+Skill generation. Use `powercontext_remember` for explicit writes; automatic capture of a run as Source evidence is
+not part of this adapter.
diff --git a/docs/en/docs/how-to/configure-openclaw.md b/docs/en/docs/how-to/configure-openclaw.md
@@ -7,7 +7,7 @@ description: Install the PowerContext memory plugin for OpenClaw and control rec
 
 ## Install or refresh the plugin
 
-Until a PowerContext release includes OpenClaw, install the CLI and plugin from the same `master` revision:
+Install the CLI and plugin from the same `master` revision:
 
 ```bash
 uv tool install --force "powercontext[cli,server] @ git+https://github.com/oceanbase/powercontext.git@master"

diff --git a/docs/en/docs/how-to/configure-pydantic-ai.md b/docs/en/docs/how-to/configure-pydantic-ai.md
@@ -1,20 +1,24 @@
 ---
-title: Configure Pydantic AI
-description: Add durable Memory tools, automatic context preparation, and optional trajectory capture to Pydantic AI.
+title: Pydantic AI adapter preview
+description: Review the current Pydantic AI adapter API and its installation status.
 ---
 
-# Configure Pydantic AI
+# Pydantic AI adapter preview
 
-Use the independently released `powercontext-pydantic-ai` package when a Pydantic AI agent should share durable
-Memory through a running PowerContext Server.
+The repository contains a preview adapter that lets a Pydantic AI agent share durable Memory through a running
+PowerContext Server. It is not yet available as a supported standalone installation.
 
-## Install the adapter
+## Check availability before using it
 
-Start the Server, then install the adapter in the agent application:
+`powercontext-pydantic-ai` is not currently published on PyPI. Its source package also requires a final
+`powercontext[client]>=0.0.3`, which the current public package and the development version from `master` do not
+satisfy. Therefore, both the old PyPI command and a direct Git subdirectory install fail dependency resolution.
 
-```bash
-uv add powercontext-pydantic-ai "pydantic-ai-slim[openai]"
-```
+Do not add this adapter to an application until compatible root and adapter packages have been released. Repository
+contributors can run its tests through the root development environment; the remaining sections document the preview
+API for development and review, not a supported installation path.
+
+## Attach the preview capability
 
 The example below uses OpenAI. For another provider, install the matching `pydantic-ai-slim` provider extra and
 change the model string.
@@ -108,5 +112,5 @@ Connecting PowerContext MCP requires no adapter package, but it is a lower-capab
 provides explicit tools; it does not automatically call `prepare_context`, capture trajectory events, or Flush at
 checkpoints and run completion.
 
-This first adapter release supports ordinary Pydantic AI runs. Durable execution through Temporal, DBOS, Prefect, or
-similar systems is not yet validated. Handoff, Candidate Review, Experience, and Skill operations are not included.
+The preview supports ordinary Pydantic AI runs. Durable execution through Temporal, DBOS, Prefect, or similar systems
+is not yet validated. Handoff, Candidate Review, Experience, and Skill operations are not included.
diff --git a/docs/en/docs/how-to/configure-workbuddy.md b/docs/en/docs/how-to/configure-workbuddy.md
@@ -7,15 +7,12 @@ description: Install the PowerContext WorkBuddy hooks and control its local beha
 
 ## Prerequisites
 
-- A running PowerContext installation. Until WorkBuddy support is included in a
-  tagged release, install the CLI and local Server from the same `master`
-  revision used below:
+- A running PowerContext installation. Install the CLI and local Server from the same `master` revision used below:
   `uv tool install "powercontext[cli,server] @ git+https://github.com/oceanbase/powercontext.git@master"`.
   Start the Server with `powercontext server run`.
 - WorkBuddy with user-level hooks, MCP, and Skills support (the desktop app).
 - Python 3.11 or newer on `PATH` for the hook process.
-- The plugin directory from this repository:
-  `integrations/workbuddy/plugins/powercontext`.
+- The plugin directory from this repository: `integrations/workbuddy/plugins/powercontext`.
 
 The integration does not start or embed the Server; it only talks to a running
 PowerContext Server over HTTP.