fix(init): correct Bedrock + LM Studio wizard schema

Bedrock prompts were producing a non-booting .env because the wizard
asked for BEDROCK_API_KEY but the config loader at
src/config/index.js:127 reads AWS_BEDROCK_API_KEY. The misleading "or
use IAM" hint also implied a SigV4 fallback that doesn't exist —
src/clients/databricks.js:1450 throws if the bearer token is missing.

LM Studio's default endpoint was http://localhost:1234/v1, but the
client at databricks.js:1329 appends /v1/chat/completions to whatever
endpoint is supplied. Result was a doubled /v1 in the request URL and
404 responses. Corrected to http://localhost:1234 to match the LM
Studio default and the maintainer's working .env.

Wizard now also collects AWS_BEDROCK_REGION and AWS_BEDROCK_MODEL_ID
when bedrock is picked, with the cross-region inference profile
(us.anthropic.claude-sonnet-4-5) as the default.

docs/init.md: corrects the providers table row and adds a Bedrock
setup subsection covering the bearer-token-vs-IAM nuance, the us./eu.
cross-region prefix, model-ID format, and tier syntax with concrete
examples for the Anthropic Claude family and Llama.

Docker image / labels bumped to 9.7.2.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: vishal veerareddy

Dockerfile

@@ -23,7 +23,7 @@ FROM node:24-alpine AS runtime
 
 ARG VCS_REF
 ARG BUILD_DATE
-ARG VERSION=9.7.1
+ARG VERSION=9.7.2
 
 LABEL org.opencontainers.image.title="Lynkr" \
       org.opencontainers.image.description="Universal LLM proxy for Claude Code, Cursor, and AI coding tools" \

bin/lynkr-init.js

@@ -55,7 +55,7 @@ const PROVIDERS = {
     local: true,
     creds: [],
     extras: [
-      { key: 'LMSTUDIO_ENDPOINT', label: 'endpoint', default: 'http://localhost:1234/v1' },
+      { key: 'LMSTUDIO_ENDPOINT', label: 'endpoint', default: 'http://localhost:1234' },
     ],
     defaultModel: 'qwen2.5-coder',
   },
@@ -112,10 +112,16 @@ const PROVIDERS = {
     label: 'AWS Bedrock',
     local: false,
     creds: [
-      { key: 'BEDROCK_API_KEY', label: 'AWS Bedrock API key (or use IAM)', secret: true },
+      // Bearer token from AWS Console → Bedrock → API Keys. The Bedrock client
+      // (src/clients/databricks.js:1450) requires this key and does NOT fall
+      // back to IAM/SigV4 — common misconception worth being explicit about.
+      { key: 'AWS_BEDROCK_API_KEY', label: 'AWS Bedrock API key (Bearer token from Bedrock console)', secret: true },
     ],
-    extras: [],
-    defaultModel: 'anthropic.claude-sonnet-4-v1:0',
+    extras: [
+      { key: 'AWS_BEDROCK_REGION', label: 'AWS region', default: 'us-east-1' },
+      { key: 'AWS_BEDROCK_MODEL_ID', label: 'Default model ID', default: 'us.anthropic.claude-sonnet-4-5-20250929-v1:0' },
+    ],
+    defaultModel: 'us.anthropic.claude-sonnet-4-5-20250929-v1:0',
   },
   vertex: {
     label: 'Google Vertex AI',

docker-compose.yml

@@ -3,7 +3,7 @@ services:
   lynkr:
     build: .
     container_name: lynkr
-    image: lynkr:9.7.1
+    image: lynkr:9.7.2
     ports:
       - "8081:8081"
     extra_hosts:
@@ -337,7 +337,7 @@ services:
       retries: 3
       start_period: 40s
     labels:
-      - "com.lynkr.version=9.7.1"
+      - "com.lynkr.version=9.7.2"
       - "com.lynkr.description=Claude Code proxy with multi-provider support"
     # Uncomment to set resource limits
     # deploy:

docs/init.md

@@ -77,13 +77,42 @@ The wizard covers everything in `src/config/index.js` `SUPPORTED_MODEL_PROVIDERS
 | `openai` | | `OPENAI_API_KEY` |
 | `openrouter` | | `OPENROUTER_API_KEY` |
 | `databricks` | | `DATABRICKS_API_BASE`, `DATABRICKS_API_KEY` |
-| `bedrock` | | `BEDROCK_API_KEY` (or IAM credentials) |
+| `bedrock` | | `AWS_BEDROCK_API_KEY` (Bearer token; no IAM fallback). See [Bedrock setup](#bedrock-setup) below. |
 | `vertex` | | `VERTEX_API_KEY` (or Application Default Credentials) |
 | `zai` | | `ZAI_API_KEY` |
 | `moonshot` | | `MOONSHOT_API_KEY` |
 
 Local providers skip the credential prompt entirely.
 
+### Bedrock setup
+
+Bedrock differs from the other cloud providers in a few ways that trip people
+up. The wizard handles all of this if you pick `bedrock` for a tier, but the
+details are:
+
+- **Authentication is Bearer-token only.** Lynkr's Bedrock client
+  (`src/clients/databricks.js:1450`) requires `AWS_BEDROCK_API_KEY` and does
+  **not** fall back to AWS IAM / SigV4 / Application Default Credentials.
+  Generate the key at *AWS Console → Bedrock → API Keys*.
+- **Region** is picked from `AWS_BEDROCK_REGION`, falling back to `AWS_REGION`,
+  then `us-east-1`.
+- **Model IDs use the `<region>.<vendor>.<model>` format.** Use the
+  cross-region inference prefix (`us.`, `eu.`, etc.) for higher availability:
+
+  ```
+  TIER_SIMPLE=bedrock:us.anthropic.claude-haiku-4-20250514-v1:0
+  TIER_MEDIUM=bedrock:us.anthropic.claude-sonnet-4-20250514-v1:0
+  TIER_COMPLEX=bedrock:us.anthropic.claude-sonnet-4-5-20250929-v1:0
+  TIER_REASONING=bedrock:us.anthropic.claude-opus-4-1-20250915-v1:0
+  ```
+
+  Non-Anthropic Bedrock models work too with the same `bedrock:<modelId>`
+  syntax — e.g. `bedrock:meta.llama3-1-70b-instruct-v1:0`.
+- **Prompt-cache injection is auto-stripped** before dispatch. Bedrock's
+  Converse API rejects `cache_control` blocks, so `normalizeBodyForConverse`
+  (databricks.js:1477) drops them. You don't have to disable prompt caching
+  globally.
+
 ---
 
 ## What ends up in `.env`

package.json

@@ -1,6 +1,6 @@
 {
   "name": "lynkr",
-  "version": "9.7.1",
+  "version": "9.7.2",
   "description": "Self-hosted LLM gateway and tier-routing proxy for Claude Code, Cursor, and Codex. Routes across Ollama, AWS Bedrock, OpenRouter, Databricks, Azure OpenAI, llama.cpp, and LM Studio with prompt caching, MCP tools, and 60-80% cost savings.",
   "main": "index.js",
   "bin": {