@averyyy/pi-client 1.0.0-piclient.1 → 1.0.1-piclient.1

package/CHANGELOG.md

@@ -2,9 +2,14 @@
 
 ### Fixed
 
+- Reported upstream Pi 1.0.0 and its full source commit in package metadata.
+- Documented remote model operations, server-side credentials, and the 64 KiB UTF-8 request limit.
+- Enabled remote prompt-cache warming with streaming and idle modes, bounded request bodies, and session usage accounting.
 - Fixed automatic update checks to compare the installed `pi-client` version instead of the upstream `pi` version.
 - Used `pi-client` in fork help and documented `web`, `send`, server configuration, global package updates, and `/reload` without changing upstream Pi help.
 - Reported upload HTTP and connection failures with the target, response metadata, and a bounded, token-redacted body excerpt instead of JSON or network stack traces.
+- Handled malformed `settings.json` when checking for the Tau Codex extension instead of crashing `pi-client web`.
+- Used `--legacy-peer-deps` for npm-global fork updates and documented installs so existing upstream Pi installs do not trigger peer override warnings for forked prerelease aliases.
 
 ### Added
 
@@ -16,11 +21,6 @@
 - `pi-client web` command that starts the client backend in Tau mirror mode on port `1838` by default.
 - `pi-client send <path>` command for chunked file and folder uploads to pi-server.
 
-### Fixed
-
-- Handled malformed `settings.json` when checking for the Tau Codex extension instead of crashing `pi-client web`.
-- Used `--legacy-peer-deps` for npm-global fork updates and documented installs so existing upstream Pi installs do not trigger peer override warnings for forked prerelease aliases.
-
 ### Changed
 
 - Updated the upstream Pi base through commit `0e6909f0`, including the latest provider and compaction fixes.

package/README.md

@@ -12,22 +12,22 @@ npm i -g --ignore-scripts --legacy-peer-deps @averyyy/pi-client
 
 ## Use
 
-Connect to the hosted server:
+Connect to your server:
 
 ```bash
-PI_SERVER_URL=https://pi.yreva.asia pi-client
+PI_SERVER_URL=https://pi-server.example.com pi-client
 ```
 
 Send one prompt and exit:
 
 ```bash
-PI_SERVER_URL=https://pi.yreva.asia pi-client -p "Say exactly: ok"
+PI_SERVER_URL=https://pi-server.example.com pi-client -p "Say exactly: ok"
 ```
 
 Send a file or folder to the server:
 
 ```bash
-PI_SERVER_URL=https://pi.yreva.asia pi-client send /path/to/file-or-folder
+PI_SERVER_URL=https://pi-server.example.com pi-client send /path/to/file-or-folder
 ```
 
 The server saves it under its configured upload directory, which defaults to `~/.pi/upload_files`.
@@ -37,7 +37,7 @@ Start the browser UI:
 ```bash
 pi-client install npm:@averyyy/pi-tau-codex
 # or: pi install npm:@averyyy/pi-tau-codex
-PI_SERVER_URL=https://pi.yreva.asia pi-client web
+PI_SERVER_URL=https://pi-server.example.com pi-client web
 ```
 
 The web command starts `pi-client` in Tau mirror mode. Install the standalone `@averyyy/pi-tau-codex` extension into the shared `~/.pi/agent` settings first. Tau listens on `http://127.0.0.1:1838` by default.
@@ -50,6 +50,24 @@ If your server uses an auth token, set it on the client:
 PI_SERVER_AUTH_TOKEN=your-token PI_SERVER_URL=http://127.0.0.1:4217 pi-client
 ```
 
+Use matching client and server releases. Provider credentials configured on the server are sufficient for chat, image generation, classifiers, and compaction. Client-side credentials, environment overrides, and request headers are forwarded when supplied.
+
+The authenticated model catalog includes chat, image, and classifier models with availability tracked separately for each model type. Extensions and codemode scripts use the normal model registry APIs; inference runs on the server. MCP connections, tools, and provider login remain in the client process.
+
+`PI_CLIENT_MAX_REQUEST_KB=64` limits each client HTTP request body to 65,536 UTF-8 bytes. Larger prompts, tool definitions, images, operation payloads, and session trees use the shared chunk transport.
+
+## Prompt Cache Warming
+
+Cache warming replays eligible requests through pi-server with the original prompt, tools, model, and cache options, limiting output to one token. The default `streaming` mode warms during active agent runs. To also warm while waiting for input, set this in your global Pi settings:
+
+```json
+{
+  "cacheWarming": "idle"
+}
+```
+
+The server confirms the cache retention used by each request, and the client uses that policy for refresh timing and cost decisions. The model must declare a prompt-cache lifetime, and the expected savings must reach $0.05. `/session` shows the warming decision. Set `cacheWarming` to `"off"` to disable it. Refresh usage counts toward session totals without adding model messages; cancellation and changed model/context stop warming. Requests use the same 64 KiB transport limit.
+
 ## Devin Pro
 
 Update both `@averyyy/pi-client` and `@averyyy/pi-server` to a version with native Devin support. Start `pi-client` with your server connection configured, then:
@@ -59,7 +77,7 @@ Update both `@averyyy/pi-client` and `@averyyy/pi-server` to a version with nati
 /model devin/swe-2-medium
 ```
 
-Login opens Devin's browser OAuth flow and saves the session token in Pi's normal credential store. The account catalog loads after login; `pi-client --list-models swe-2` lists the discovered SWE-2 variants. Choose the exact variant (for example `swe-2-high` or `swe-2-max`) to change effort. Available models and limits come from Devin; there is no hardcoded fallback catalog.
+Login opens Devin's browser OAuth flow and saves the session token in Pi's normal credential store. The account catalog loads after login; `pi-client --list-models swe-2` lists the discovered SWE-2 variants. Choose the exact variant (for example `swe-2-high` or `swe-2-max`) to change effort. Known configured SWE-2 IDs remain selectable before account discovery finishes; successful discovery supplies the account's models and limits, and discovery failures are reported.
 
 Authentication and model discovery may contact Devin from the client. Model inference, including compaction, goes through `pi-client → pi-server → Devin`. Direct Devin inference rejects `PI_SERVER_MODE=true`. Pi continues to own tools, agent loops, session history, and subagents. No Devin CLI installation or provider extension is needed. Remove conflicting Devin extensions before enabling the native provider.
 

package/package.json

@@ -1,11 +1,11 @@
 {
 	"name": "@averyyy/pi-client",
-	"version": "1.0.0-piclient.1",
+	"version": "1.0.1-piclient.1",
 	"description": "Lightweight CLI wrapper that connects to a pi-server instance",
 	"type": "module",
 	"piClient": {
-		"basePiVersion": "0.80.6",
-		"basePiCommit": "0e6909f0"
+		"basePiVersion": "1.0.1",
+		"basePiCommit": "a13d35a742c6ef8462812a28fbe1d8c8b7431c32"
 	},
 	"bin": {
 		"pi-client": "bin/pi-client.js"
@@ -25,7 +25,7 @@
 		"prepublishOnly": "npm run build"
 	},
 	"dependencies": {
-		"@earendil-works/pi-coding-agent": "npm:@averyyy/pi-coding-agent@1.0.0-piclient.1"
+		"@earendil-works/pi-coding-agent": "npm:@averyyy/pi-coding-agent@1.0.1-piclient.1"
 	},
 	"devDependencies": {
 		"shx": "0.4.0",