localpi 0.6.1 → 0.6.3

package/README.md

@@ -109,6 +109,10 @@ Localpi launches Pi with:
 - in-session `/thinking` (Pi's own command) and `/approval` (localpi's) for changing session settings
 - local state under `~/.local/state/localpi`
 
+Pi owns stdout. Localpi writes its own diagnostics, including the connection summary and catalog
+warnings, to stderr. A machine-readable Pi mode such as `--mode rpc` or `--mode json` therefore stays
+parseable, and a batch run keeps a clean stdout.
+
 The approval gate makes failed or denied tool calls explicit to the model so the model does not claim that a blocked command ran.
 
 ## Skills
@@ -524,6 +528,7 @@ demowall record --session demowall-<timestamp> --out demo.mp4 --seconds 60
 - `--max-tokens <n>`: generated model max output tokens
 - `--continue-on-truncation <n>`: continue a reply cut off by the output limit, up to `n` times. Off by default, and `LOCALPI_CONTINUE_ON_TRUNCATION=<n>` sets the same limit
 - `--base-url <url>`: OpenAI-compatible endpoint for LM Studio or custom endpoints
+- `--api-key <value>`: Pi provider API key for this run. A literal value, an environment reference such as `${NAME}`, or a `!command`. Default: `local`. `LOCALPI_API_KEY` sets the same value
 - `--server-command <path>`: `llama-server` executable path
 - `--llama-server <path>`: alias for `--server-command`
 - `--host <host>`: managed `llama-server` host. Default: `127.0.0.1`
@@ -583,6 +588,7 @@ explicit `PI_OFFLINE=0` or `PI_OFFLINE=1` always wins.
 - `LOCALPI_MODEL`
 - `LOCALPI_PROVIDER`
 - `LOCALPI_BASE_URL`
+- `LOCALPI_API_KEY`
 - `LOCALPI_PROVIDERS_FILE`
 - `LOCALPI_MODEL_PROFILE`
 - `LOCALPI_MODEL_REASONING`

package/dist/src/cli/cli.js

@@ -307,7 +307,13 @@ async function launchResolvedRuntime(app, connection) {
     if (code !== 0) {
         return { code, stdout: "", stderr: "" };
     }
-    return ok(connection.warnings.length === 0 ? "" : connectionStatus(connection));
+    // Pi owns stdout. Localpi reports its own diagnostics on stderr, so a machine-readable Pi mode
+    // such as --mode rpc or --mode json stays parseable.
+    return {
+        code: 0,
+        stdout: "",
+        stderr: connection.warnings.length === 0 ? "" : connectionStatus(connection)
+    };
 }
 /**
  * ACP mode keeps stdout for the protocol only, so localpi reports its own diagnostics on stderr and

package/dist/src/localpi/options.js

@@ -19,6 +19,7 @@ export function defaultOptions() {
     return {
         runtime: parseRuntime(envString("LOCALPI_RUNTIME", "auto")),
         baseUrl: envOptionalBaseUrl("LOCALPI_BASE_URL"),
+        apiKey: envString("LOCALPI_API_KEY", "local"),
         model: process.env["LOCALPI_MODEL"],
         provider: process.env["LOCALPI_PROVIDER"],
         customProviderId: envString("LOCALPI_PROVIDER_ID", "local-openai"),
@@ -100,6 +101,8 @@ export function usage() {
         "  --provider <id>          catalog provider id to use",
         "  --model <alias|id|path>  model alias, backend id, or GGUF path",
         "  --base-url <url>         OpenAI-compatible endpoint",
+        "  --api-key <value>        Pi provider API key: a literal, ${NAME}, or !command",
+        "                          (LOCALPI_API_KEY=<value>, default: local)",
         "  --ctx <n>                model context window",
         "  --context-window <n>     alias for --ctx",
         "  --max-tokens <n>         generated model max output tokens",
@@ -189,6 +192,7 @@ const booleanFlagUpdaters = {
 const valueFlagUpdaters = {
     "--runtime": (options, value) => ({ ...options, runtime: parseRuntime(value) }),
     "--base-url": (options, value) => ({ ...options, baseUrl: normalizeBaseUrl(value) }),
+    "--api-key": (options, value) => ({ ...options, apiKey: value }),
     "--model": (options, value) => ({ ...options, model: value }),
     "--provider": (options, value) => ({ ...options, provider: value }),
     "--provider-id": (options, value) => ({ ...options, customProviderId: value }),

package/dist/src/pi/app.js

@@ -86,7 +86,7 @@ function providersForConnection(options, connection) {
         id: group.providerId,
         baseUrl: group.baseUrl,
         api: "openai-completions",
-        apiKey: "local",
+        apiKey: options.apiKey,
         compat: {
             supportsDeveloperRole: false,
             supportsReasoningEffort: false

package/docs/runtime-specification.md

@@ -110,6 +110,7 @@ Localpi:
 - can use `--provider <id>` to name the generated Pi provider
 - uses `/v1/models` for discovery
 - avoids assuming it can start, stop, or unload the backend
+- writes the `--api-key` value as the generated Pi provider key, exactly as given, so an environment reference such as `${HF_TOKEN}` or a `!command` keeps the secret out of the Pi config. The default is `local`, which local engines accept and ignore
 
 ### Configured Providers
 
@@ -200,6 +201,10 @@ Localpi installs two default extensions:
 - tool approval gate: ask before each tool call, and tell the model clearly when a tool call was blocked
 - token status: show live generation speed while streaming, then final prefill and generation rates when usage data is available
 
+Pi owns stdout. Localpi writes its own diagnostics, including the connection summary
+and catalog warnings, to stderr, so a machine-readable Pi mode such as `--mode rpc`
+or `--mode json` stays parseable and a batch run keeps a clean stdout.
+
 ## System Prompt
 
 Localpi appends a short system prompt that tells the model:

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "localpi",
-  "version": "0.6.1",
+  "version": "0.6.3",
   "description": "Swiss army knife for running Pi with local inference engines.",
   "type": "module",
   "license": "MIT",