cascade-ai 0.69.0 → 0.70.0

package/README.md

@@ -43,6 +43,8 @@ Other AI CLIs run a single agent. Cascade runs a visible **organization** — an
 - [Features](#features)
 - [Installation](#installation)
 - [Quick Start](#quick-start)
+- [Self-host](#self-host)
+- [OpenAI-compatible API](#openai-compatible-api)
 - [Configuration](#configuration)
 - [AI Providers](#ai-providers)
 - [Tools](#tools)
@@ -246,6 +248,39 @@ See the [Dockerfile](Dockerfile) and [docker-compose.yml](docker-compose.yml) fo
 
 ---
 
+## OpenAI-compatible API
+
+Anything that already talks to OpenAI can talk to Cascade. Point the client's `base_url` at your server's `/v1` and use a Cascade access token as the API key — `POST /v1/chat/completions` and `GET /v1/models` work with the official SDKs, streaming and not.
+
+```python
+from openai import OpenAI
+
+client = OpenAI(api_key=CASCADE_ACCESS_TOKEN, base_url="http://localhost:8787/v1")
+
+reply = client.chat.completions.create(
+    model="cascade",                                    # a routing mode, not a model
+    messages=[{"role": "user", "content": "Compare Postgres and SQLite for a CLI tool."}],
+)
+print(reply.choices[0].message.content)
+print(reply.cascade)   # which tier + model actually served it, and what routing saved
+```
+
+`model` names a **routing mode**, because Cascade picks a model per subtask — that is the product:
+
+| `model` | what runs |
+| --- | --- |
+| `cascade` | full orchestration, balanced quality against cost |
+| `cascade-fast` | one mid-tier model, no orchestration |
+| `cascade-quality` | full orchestration, biased to quality |
+
+Anything else returns `404 model_not_found` rather than quietly running something you didn't ask for. Unsupported parameters (`n > 1`, `logprobs`, `tools`, `response_format`, …) are **rejected**, not ignored — a silently dropped parameter returns a response that looks successful and is wrong. `temperature` and `max_tokens` are honoured, applied across tiers.
+
+**Provider keys.** On a **single-account** instance — a self-host, where the operator and the caller are the same person — the endpoint uses the provider keys in your `.env` (`OPENAI_API_KEY`, `ANTHROPIC_API_KEY`, …; the same names the CLI reads). The moment a second account exists this stops automatically, because the operator's key would otherwise pay for everyone else's runs. Any instance can also take keys per request via the SDK's `extra_body={"providers": [...]}`.
+
+`/v1` is for server-side clients: `Authorization` is not allowed cross-origin, so a browser-side SDK is deliberately not served. Tools/function calling and image inputs are not in v1 — Cascade's tools run server-side, and attachments go through `POST /api/uploads`.
+
+---
+
 ## Configuration
 
 Cascade loads config from `.cascade/config.json` in your project directory.

package/dist/cli.cjs

@@ -1810,7 +1810,7 @@ var CASCADE_VERSION, CASCADE_CONFIG_FILE, CASCADE_DB_FILE, CASCADE_DASHBOARD_SEC
 var init_constants = __esm({
   "src/constants.ts"() {
     init_pricing();
-    CASCADE_VERSION = "0.69.0";
+    CASCADE_VERSION = "0.70.0";
     CASCADE_CONFIG_FILE = ".cascade/config.json";
     CASCADE_DB_FILE = ".cascade/memory.db";
     CASCADE_DASHBOARD_SECRET_FILE = ".cascade/dashboard-secret";

package/dist/cli.js

@@ -1757,7 +1757,7 @@ var CASCADE_VERSION, CASCADE_CONFIG_FILE, CASCADE_DB_FILE, CASCADE_DASHBOARD_SEC
 var init_constants = __esm({
   "src/constants.ts"() {
     init_pricing();
-    CASCADE_VERSION = "0.69.0";
+    CASCADE_VERSION = "0.70.0";
     CASCADE_CONFIG_FILE = ".cascade/config.json";
     CASCADE_DB_FILE = ".cascade/memory.db";
     CASCADE_DASHBOARD_SECRET_FILE = ".cascade/dashboard-secret";

package/dist/desktop-core.cjs

@@ -291781,7 +291781,7 @@ function reconcilePrice(model, live) {
 }
 
 // src/constants.ts
-var CASCADE_VERSION = "0.69.0";
+var CASCADE_VERSION = "0.70.0";
 var CASCADE_CONFIG_DIR = ".cascade";
 var CASCADE_MD_FILE = "CASCADE.md";
 var CASCADE_IGNORE_FILE = ".cascadeignore";

package/dist/index.cjs

@@ -1908,7 +1908,7 @@ function reconcilePrice(model, live) {
 }
 
 // src/constants.ts
-var CASCADE_VERSION = "0.69.0";
+var CASCADE_VERSION = "0.70.0";
 var CASCADE_CONFIG_DIR = ".cascade";
 var CASCADE_MD_FILE = "CASCADE.md";
 var CASCADE_IGNORE_FILE = ".cascadeignore";

package/dist/index.js

@@ -1862,7 +1862,7 @@ function reconcilePrice(model, live) {
 }
 
 // src/constants.ts
-var CASCADE_VERSION = "0.69.0";
+var CASCADE_VERSION = "0.70.0";
 var CASCADE_CONFIG_DIR = ".cascade";
 var CASCADE_MD_FILE = "CASCADE.md";
 var CASCADE_IGNORE_FILE = ".cascadeignore";

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cascade-ai",
-  "version": "0.69.0",
+  "version": "0.70.0",
   "description": "Multi-tier AI orchestration CLI — T1/T2/T3 hierarchical agent system",
   "type": "module",
   "main": "dist/index.cjs",