cade-cli 0.13.4__tar.gz → 0.14.0__tar.gz

{cade_cli-0.13.4 → cade_cli-0.14.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: cade-cli
-Version: 0.13.4
+Version: 0.14.0
 Summary: Cade - The CLI Agent from Arcade.dev
 Project-URL: Homepage, https://arcade.dev
 Project-URL: Documentation, https://docs.arcade.dev
@@ -91,7 +91,7 @@ Description-Content-Type: text/markdown
 The CLI agent from [Arcade.dev](https://arcade.dev). Coding, research, and everyday automation — driven from a terminal, a single shell pipe, or a long-lived daemon that replies over Telegram.
 
 ```bash
-brew install ArcadeAI/tap/cade   #  or:  uv tool install cade-cli   |   pip install cade-cli
+curl -fsSL https://arcadeagent.dev/install | sh   #  or:  uv tool install cade-cli  |  pip install cade-cli
 cade login
 cade
 ```
@@ -101,10 +101,10 @@ cade
 ## Highlights
 
 - **Three deployment modes from the same binary.** Interactive chat (`cade`), one-shot or pipe (`cade -m "…"`), or a headless daemon that replies over messaging (`cade serve`).
-- **Tools from anywhere.** Built-in local tools (filesystem / shell / git / search / tasks / channels / memory), [Arcade Cloud](https://arcade.dev) tools (Slack, GitHub, Gmail, calendars, …), and any [MCP](https://modelcontextprotocol.io/) server you register — all unified under one tool surface.
+- **Tools from anywhere.** Built-in local tools (filesystem / shell / git / search / tasks / memory), [Arcade Cloud](https://arcade.dev) tools (Slack, GitHub, Gmail, calendars, …), and any [MCP](https://modelcontextprotocol.io/) server you register — all unified under one tool surface.
 - **Bidirectional MCP.** Tools can call `context.ui.elicit(…)` for structured input, `context.log.*` for live operator logs, and `context.progress.report(…)` for editable progress UI — routed end-to-end through whichever adapter is active (Telegram inline keyboard, terminal prompt).
-- **Personas + per-persona memory.** Multiple agent profiles with their own system prompt and an indexed memory store backed by [agent-library](https://pypi.org/project/agent-library/).
-- **Extensibility surface.** Hooks (lifecycle events), tasks (file-backed durable work), cron (scheduled prompts), channels (cross-session pub/sub) — all available to the agent as `Local_*` tools and to you as `cade <subcommand>`.
+- **Agents + per-agent memory.** Multiple agents, each with their own model, system prompt, tools, and an indexed memory store backed by [agent-library](https://pypi.org/project/agent-library/).
+- **Extensibility surface.** Hooks (lifecycle events), tasks (durable work), cron (scheduled prompts) — all available to the agent as `Local_*` tools and to you as `cade <subcommand>`.
 - **Bring your own LLM.** OpenAI, Anthropic, or any OpenAI-compatible endpoint (Ollama, vLLM, Together, Groq, Fireworks, …). `--local-only` skips Arcade Cloud entirely.
 
 ---
@@ -113,12 +113,21 @@ cade
 
 | Source | Command |
 |---|---|
-| Homebrew | `brew install ArcadeAI/tap/cade` |
+| **Install script** (recommended) | `curl -fsSL https://arcadeagent.dev/install \| sh` |
 | uv | `uv tool install cade-cli` |
 | pip | `pip install cade-cli` |
 | From source | `git clone https://github.com/arcadeai-labs/cade.git && cd cade && uv sync` |
 
-Homebrew installs from a private release feed. Export `HOMEBREW_GITHUB_API_TOKEN` (a token with read access to `arcadeai-labs/cade`) before running `brew install`.
+The install script detects your OS/arch, downloads the matching prebuilt binary
+(macOS arm64, Linux x64), verifies its checksum, and drops `cade` in
+`~/.local/bin` — no GitHub token required. Pin a version or install location:
+
+```bash
+CADE_VERSION=0.14.0 CADE_INSTALL_DIR=/usr/local/bin \
+  sh -c "$(curl -fsSL https://arcadeagent.dev/install)"
+```
+
+(Windows: grab the `.zip` from the releases page, or `pip install cade-cli`.)
 
 **Prerequisites:** Python 3.11+, an Arcade account ([arcade.dev](https://arcade.dev)) for cloud tools, and an LLM provider key (`OPENAI_API_KEY` or `ANTHROPIC_API_KEY`). Skip the Arcade account with `--local-only` or `CADE_LOCAL_ONLY=1`.
 
@@ -138,9 +147,9 @@ Full install and first-run docs: [`docs/install.md`](./docs/install.md) and
 cade                                       # interactive chat
 cade -m "What changed in HEAD?"            # single message, then exit
 cade -m "What went wrong?" < error.log     # prompt + file/stdin
-cade -r                                    # resume the most recent thread
+cade resume                                # resume the most recent thread
 cade resume "auth-rewrite"                 # resume a thread by name
-cade --persona reviewer                    # use a saved persona's system prompt
+cade use reviewer                          # switch the current agent
 cade --voice                               # speak / hear (requires cade-cli[voice])
 ```
 
@@ -148,10 +157,8 @@ cade --voice                               # speak / hear (requires cade-cli[voi
 
 | Flag | Description |
 |---|---|
-| `-r`, `--resume` | Resume most recent thread |
 | `-m`, `--message` | Single-message mode; piped stdin is appended to the prompt |
-| `-P`, `--persona` | Use a saved persona |
-| `-V`, `--voice` | Voice mode; install `cade-cli[voice]` first |
+| `--voice` | Voice mode; install `cade-cli[voice]` first |
 | `-v`, `--verbose` | Debug logging |
 | `--version` | Print version |
 
@@ -165,7 +172,7 @@ cade --voice                               # speak / hear (requires cade-cli[voi
 | `/logs` | Recent log entries |
 | `/thread`, `/history` | Current thread info |
 | `/pin`, `/unpin` | Pin reference material into the session |
-| `/tasks`, `/cron`, `/hooks`, `/channels`, `/notify` | Coordination surfaces (mirrored as `cade tasks` / `cade cron` / `cade hooks` / `cade channels`) |
+| `/tasks`, `/cron`, `/hooks`, `/notify` | Coordination surfaces (mirrored as `cade tasks` / `cade cron` / `cade hooks`) |
 | `/usage` | Context-window status and token usage |
 | `/cd`, `/pwd`, `/!` | Shell shortcuts |
 | `Ctrl+C` | Exit |
@@ -177,7 +184,7 @@ cade --voice                               # speak / hear (requires cade-cli[voi
 Tools come from three sources and are addressed uniformly by their registered name:
 
 - **`Local_*`** — built-in: filesystem (`ReadFile`, `WriteFile`, `Edit`, `ListFiles`), shell (`Bash`), `Search`, `Git`, `Task*`, `AskUserQuestion`, `RetrieveToolResult`, `ToolSchema`, `PinContext`, `PushNotification`, `WebFetch`, and `WebSearch`.
-- **`Memory_*`** — agent-library backed, per-persona indexed knowledge store.
+- **`Memory_*`** — agent-library backed, per-agent indexed knowledge store.
 - **Arcade Cloud + user MCP servers** — anything you register.
 
 ```bash
@@ -198,37 +205,46 @@ cade mcp authorize my-server                      # OAuth flow (browser)
 cade mcp list / status / test / enable / disable / rm
 ```
 
-User MCP servers register at `~/.cade/config/contexts/<context>.toml [[mcp]]`. Servers with large catalogs are auto-deferred (parameter schemas stripped to keep tool-list bytes down) and fetched on demand via `Local_ToolSchema`.
+User MCP servers register on the agent's `~/.cade/config/agents/<name>.toml [[mcp]]` list (with `cade.toml [agent_defaults].mcp` as the shared default). Servers with large catalogs are auto-deferred (parameter schemas stripped to keep tool-list bytes down) and fetched on demand via `Local_ToolSchema`.
 
 ---
 
-## Personas + memory
+## Agents + memory
+
+Each **agent** bundles a model, system prompt, tools, and memory. The current
+agent is ambient — most commands act on it.
 
 ```bash
-cade persona list / create / get / edit / delete / default   # manage system prompts
-cade --persona reviewer                       # use one for a session
+cade new reviewer --prompt "You are a meticulous code reviewer."  # create + switch
+cade list                                     # all agents
+cade use reviewer                             # switch the current agent
+cade info / cade edit reviewer                # inspect / change config
 
-cade mem list                                 # sources indexed for the active persona
+cade mem                                      # the current agent's memory (status)
 cade mem add ~/notes                          # index a directory
 cade mem search "auth design"                 # semantic search
 ```
 
-Memory is persona-scoped — each persona has its own `~/.cade/data/contexts/<context>/memory/<persona>/.librarian/` index, so a `reviewer` persona's notes never leak into `default`.
+Memory is per-agent — each agent has its own
+`~/.cade/data/agents/<id>/memory/.librarian/` index, so a `reviewer` agent's
+notes never leak into another.
 
 ---
 
 ## Headless daemon: `cade serve`
 
-Run Cade as a long-lived daemon that replies through adapters. Telegram is supported for personal bot workflows, and Desktop exposes a local WebSocket for native clients. Each conversation gets its own persistent thread — persona, memory, and tools carry across messages.
+Run Cade as a long-lived daemon that replies through adapters. Telegram is supported for personal bot workflows, and Desktop exposes a local WebSocket for native clients. Each conversation gets its own persistent thread — model, memory, and tools carry across messages.
 
 ```bash
 cade serve init                          # wizard: bot token, allowed senders
-cade serve                               # foreground
+cade serve                               # foreground (current agent)
+cade serve --agent ops                   # serve a specific agent (one daemon per agent)
+cade serve list                          # agents with a running daemon
 cade serve --install launchd             # macOS LaunchAgent (or --install systemd)
 cade serve status / stop                 # health, uptime; SIGTERM running daemon
 ```
 
-Config lives at `~/.cade/config/contexts/<context>.toml [serve]`. Edit and `kill -HUP $(cat ~/.cade/run/serve.pid)` to reload without restart.
+Each agent's serve config lives in `~/.cade/config/agents/<name>.toml [serve]`. Edit and `kill -HUP $(cat ~/.cade/run/agents/<name>/serve.pid)` to reload without restart.
 
 **Security defaults are fail-closed:** `allowed_senders = []` blocks everyone; tools are default-deny against `[serve.tools].allow` (globs / regex / pipe lists supported); `deny` always wins.
 
@@ -236,18 +252,19 @@ The MCP transport is fully bidirectional, so tools running inside the daemon can
 
 ---
 
-## Extensibility: hooks, tasks, cron, channels
+## Extensibility: hooks, tasks, cron
 
-Four primitives the agent uses directly (as `Local_*` tools) and you can drive from the CLI:
+Three primitives the agent uses directly (as `Local_*` tools) and you can drive from the CLI:
 
 | Surface | What it is | CLI |
 |---|---|---|
 | **Hooks** | Lifecycle event handlers (pre/post tool, user prompt submit) | `cade hooks list / add / remove / test` |
-| **Tasks** | File-backed durable work units that survive turn boundaries | `cade tasks list / get / create / update / delete` |
+| **Tasks** | Durable work units that survive turn boundaries | `cade tasks list / get / create / update / delete` |
 | **Cron** | Scheduled prompts the agent runs on a timer | `cade cron list / add / remove / enable / disable / run-now` |
-| **Channels** | JSONL-backed inter-session pub/sub | `cade channels list / send / recv / clear` |
 
-See [`docs/channels.md`](./docs/channels.md) for the channels protocol; hooks and tasks/cron details live in their respective `cade <subcommand> --help` outputs.
+Cross-agent messaging is set up with `cade link` / `cade channel` (an agent can only
+reach another it's been linked to). Hooks and tasks/cron details live in their
+respective `cade <subcommand> --help` outputs.
 
 ---
 
@@ -259,9 +276,9 @@ Cade works with any OpenAI-compatible endpoint. Three ways to wire one in (CLI f
 # CLI flag for Ollama. Use a model from `ollama list`.
 cade chat -L --endpoint http://localhost:11434 --model qwen3:4b
 
-# Environment for one-shot mode
-cade context create local-qwen --model qwen3:4b
-cade context use local-qwen
+# Or create a dedicated agent for the local model
+cade new local-qwen --model qwen3:4b
+cade use local-qwen
 export OLLAMA_BASE_URL="http://localhost:11434"
 CADE_LOCAL_ONLY=1 cade -m "hello from a local model"
 
@@ -281,12 +298,11 @@ All state lives under `~/.cade/`. Override the location with `CADE_HOME=/path/to
 
 | File | Purpose |
 |---|---|
-| `config/cade.toml` | Model / provider / UI settings |
-| `data/contexts/<context>/history.db` | Thread + message storage |
+| `config/cade.toml` | Global settings + `[agent_defaults]` (model, provider, UI, …) |
+| `config/agents/<name>.toml` | Per-agent config (model, prompt, `[[mcp]]`, `[serve]`, hooks) |
+| `data/agents/<id>/` | Per-agent state: `threads/ memory/ tasks/ inbox/` |
+| `run/agents/<name>/` | Per-agent daemon runtime: `serve.pid`, mcp-services, cron lease |
 | `logs/cade.log` | Rotating log file |
-| `config/contexts/<context>.toml [[mcp]]` | User MCP server registry |
-| `config/contexts/<context>.toml [serve]` | `cade serve` daemon config |
-| `personas/`, `memory/`, `tasks/`, `channels/` | Persona-scoped state |
 
 Full layout, env vars, and TOML schema in [`docs/configuration.md`](./docs/configuration.md).
 

{cade_cli-0.13.4 → cade_cli-0.14.0}/README.md

@@ -3,7 +3,7 @@
 The CLI agent from [Arcade.dev](https://arcade.dev). Coding, research, and everyday automation — driven from a terminal, a single shell pipe, or a long-lived daemon that replies over Telegram.
 
 ```bash
-brew install ArcadeAI/tap/cade   #  or:  uv tool install cade-cli   |   pip install cade-cli
+curl -fsSL https://arcadeagent.dev/install | sh   #  or:  uv tool install cade-cli  |  pip install cade-cli
 cade login
 cade
 ```
@@ -13,10 +13,10 @@ cade
 ## Highlights
 
 - **Three deployment modes from the same binary.** Interactive chat (`cade`), one-shot or pipe (`cade -m "…"`), or a headless daemon that replies over messaging (`cade serve`).
-- **Tools from anywhere.** Built-in local tools (filesystem / shell / git / search / tasks / channels / memory), [Arcade Cloud](https://arcade.dev) tools (Slack, GitHub, Gmail, calendars, …), and any [MCP](https://modelcontextprotocol.io/) server you register — all unified under one tool surface.
+- **Tools from anywhere.** Built-in local tools (filesystem / shell / git / search / tasks / memory), [Arcade Cloud](https://arcade.dev) tools (Slack, GitHub, Gmail, calendars, …), and any [MCP](https://modelcontextprotocol.io/) server you register — all unified under one tool surface.
 - **Bidirectional MCP.** Tools can call `context.ui.elicit(…)` for structured input, `context.log.*` for live operator logs, and `context.progress.report(…)` for editable progress UI — routed end-to-end through whichever adapter is active (Telegram inline keyboard, terminal prompt).
-- **Personas + per-persona memory.** Multiple agent profiles with their own system prompt and an indexed memory store backed by [agent-library](https://pypi.org/project/agent-library/).
-- **Extensibility surface.** Hooks (lifecycle events), tasks (file-backed durable work), cron (scheduled prompts), channels (cross-session pub/sub) — all available to the agent as `Local_*` tools and to you as `cade <subcommand>`.
+- **Agents + per-agent memory.** Multiple agents, each with their own model, system prompt, tools, and an indexed memory store backed by [agent-library](https://pypi.org/project/agent-library/).
+- **Extensibility surface.** Hooks (lifecycle events), tasks (durable work), cron (scheduled prompts) — all available to the agent as `Local_*` tools and to you as `cade <subcommand>`.
 - **Bring your own LLM.** OpenAI, Anthropic, or any OpenAI-compatible endpoint (Ollama, vLLM, Together, Groq, Fireworks, …). `--local-only` skips Arcade Cloud entirely.
 
 ---
@@ -25,12 +25,21 @@ cade
 
 | Source | Command |
 |---|---|
-| Homebrew | `brew install ArcadeAI/tap/cade` |
+| **Install script** (recommended) | `curl -fsSL https://arcadeagent.dev/install \| sh` |
 | uv | `uv tool install cade-cli` |
 | pip | `pip install cade-cli` |
 | From source | `git clone https://github.com/arcadeai-labs/cade.git && cd cade && uv sync` |
 
-Homebrew installs from a private release feed. Export `HOMEBREW_GITHUB_API_TOKEN` (a token with read access to `arcadeai-labs/cade`) before running `brew install`.
+The install script detects your OS/arch, downloads the matching prebuilt binary
+(macOS arm64, Linux x64), verifies its checksum, and drops `cade` in
+`~/.local/bin` — no GitHub token required. Pin a version or install location:
+
+```bash
+CADE_VERSION=0.14.0 CADE_INSTALL_DIR=/usr/local/bin \
+  sh -c "$(curl -fsSL https://arcadeagent.dev/install)"
+```
+
+(Windows: grab the `.zip` from the releases page, or `pip install cade-cli`.)
 
 **Prerequisites:** Python 3.11+, an Arcade account ([arcade.dev](https://arcade.dev)) for cloud tools, and an LLM provider key (`OPENAI_API_KEY` or `ANTHROPIC_API_KEY`). Skip the Arcade account with `--local-only` or `CADE_LOCAL_ONLY=1`.
 
@@ -50,9 +59,9 @@ Full install and first-run docs: [`docs/install.md`](./docs/install.md) and
 cade                                       # interactive chat
 cade -m "What changed in HEAD?"            # single message, then exit
 cade -m "What went wrong?" < error.log     # prompt + file/stdin
-cade -r                                    # resume the most recent thread
+cade resume                                # resume the most recent thread
 cade resume "auth-rewrite"                 # resume a thread by name
-cade --persona reviewer                    # use a saved persona's system prompt
+cade use reviewer                          # switch the current agent
 cade --voice                               # speak / hear (requires cade-cli[voice])
 ```
 
@@ -60,10 +69,8 @@ cade --voice                               # speak / hear (requires cade-cli[voi
 
 | Flag | Description |
 |---|---|
-| `-r`, `--resume` | Resume most recent thread |
 | `-m`, `--message` | Single-message mode; piped stdin is appended to the prompt |
-| `-P`, `--persona` | Use a saved persona |
-| `-V`, `--voice` | Voice mode; install `cade-cli[voice]` first |
+| `--voice` | Voice mode; install `cade-cli[voice]` first |
 | `-v`, `--verbose` | Debug logging |
 | `--version` | Print version |
 
@@ -77,7 +84,7 @@ cade --voice                               # speak / hear (requires cade-cli[voi
 | `/logs` | Recent log entries |
 | `/thread`, `/history` | Current thread info |
 | `/pin`, `/unpin` | Pin reference material into the session |
-| `/tasks`, `/cron`, `/hooks`, `/channels`, `/notify` | Coordination surfaces (mirrored as `cade tasks` / `cade cron` / `cade hooks` / `cade channels`) |
+| `/tasks`, `/cron`, `/hooks`, `/notify` | Coordination surfaces (mirrored as `cade tasks` / `cade cron` / `cade hooks`) |
 | `/usage` | Context-window status and token usage |
 | `/cd`, `/pwd`, `/!` | Shell shortcuts |
 | `Ctrl+C` | Exit |
@@ -89,7 +96,7 @@ cade --voice                               # speak / hear (requires cade-cli[voi
 Tools come from three sources and are addressed uniformly by their registered name:
 
 - **`Local_*`** — built-in: filesystem (`ReadFile`, `WriteFile`, `Edit`, `ListFiles`), shell (`Bash`), `Search`, `Git`, `Task*`, `AskUserQuestion`, `RetrieveToolResult`, `ToolSchema`, `PinContext`, `PushNotification`, `WebFetch`, and `WebSearch`.
-- **`Memory_*`** — agent-library backed, per-persona indexed knowledge store.
+- **`Memory_*`** — agent-library backed, per-agent indexed knowledge store.
 - **Arcade Cloud + user MCP servers** — anything you register.
 
 ```bash
@@ -110,37 +117,46 @@ cade mcp authorize my-server                      # OAuth flow (browser)
 cade mcp list / status / test / enable / disable / rm
 ```
 
-User MCP servers register at `~/.cade/config/contexts/<context>.toml [[mcp]]`. Servers with large catalogs are auto-deferred (parameter schemas stripped to keep tool-list bytes down) and fetched on demand via `Local_ToolSchema`.
+User MCP servers register on the agent's `~/.cade/config/agents/<name>.toml [[mcp]]` list (with `cade.toml [agent_defaults].mcp` as the shared default). Servers with large catalogs are auto-deferred (parameter schemas stripped to keep tool-list bytes down) and fetched on demand via `Local_ToolSchema`.
 
 ---
 
-## Personas + memory
+## Agents + memory
+
+Each **agent** bundles a model, system prompt, tools, and memory. The current
+agent is ambient — most commands act on it.
 
 ```bash
-cade persona list / create / get / edit / delete / default   # manage system prompts
-cade --persona reviewer                       # use one for a session
+cade new reviewer --prompt "You are a meticulous code reviewer."  # create + switch
+cade list                                     # all agents
+cade use reviewer                             # switch the current agent
+cade info / cade edit reviewer                # inspect / change config
 
-cade mem list                                 # sources indexed for the active persona
+cade mem                                      # the current agent's memory (status)
 cade mem add ~/notes                          # index a directory
 cade mem search "auth design"                 # semantic search
 ```
 
-Memory is persona-scoped — each persona has its own `~/.cade/data/contexts/<context>/memory/<persona>/.librarian/` index, so a `reviewer` persona's notes never leak into `default`.
+Memory is per-agent — each agent has its own
+`~/.cade/data/agents/<id>/memory/.librarian/` index, so a `reviewer` agent's
+notes never leak into another.
 
 ---
 
 ## Headless daemon: `cade serve`
 
-Run Cade as a long-lived daemon that replies through adapters. Telegram is supported for personal bot workflows, and Desktop exposes a local WebSocket for native clients. Each conversation gets its own persistent thread — persona, memory, and tools carry across messages.
+Run Cade as a long-lived daemon that replies through adapters. Telegram is supported for personal bot workflows, and Desktop exposes a local WebSocket for native clients. Each conversation gets its own persistent thread — model, memory, and tools carry across messages.
 
 ```bash
 cade serve init                          # wizard: bot token, allowed senders
-cade serve                               # foreground
+cade serve                               # foreground (current agent)
+cade serve --agent ops                   # serve a specific agent (one daemon per agent)
+cade serve list                          # agents with a running daemon
 cade serve --install launchd             # macOS LaunchAgent (or --install systemd)
 cade serve status / stop                 # health, uptime; SIGTERM running daemon
 ```
 
-Config lives at `~/.cade/config/contexts/<context>.toml [serve]`. Edit and `kill -HUP $(cat ~/.cade/run/serve.pid)` to reload without restart.
+Each agent's serve config lives in `~/.cade/config/agents/<name>.toml [serve]`. Edit and `kill -HUP $(cat ~/.cade/run/agents/<name>/serve.pid)` to reload without restart.
 
 **Security defaults are fail-closed:** `allowed_senders = []` blocks everyone; tools are default-deny against `[serve.tools].allow` (globs / regex / pipe lists supported); `deny` always wins.
 
@@ -148,18 +164,19 @@ The MCP transport is fully bidirectional, so tools running inside the daemon can
 
 ---
 
-## Extensibility: hooks, tasks, cron, channels
+## Extensibility: hooks, tasks, cron
 
-Four primitives the agent uses directly (as `Local_*` tools) and you can drive from the CLI:
+Three primitives the agent uses directly (as `Local_*` tools) and you can drive from the CLI:
 
 | Surface | What it is | CLI |
 |---|---|---|
 | **Hooks** | Lifecycle event handlers (pre/post tool, user prompt submit) | `cade hooks list / add / remove / test` |
-| **Tasks** | File-backed durable work units that survive turn boundaries | `cade tasks list / get / create / update / delete` |
+| **Tasks** | Durable work units that survive turn boundaries | `cade tasks list / get / create / update / delete` |
 | **Cron** | Scheduled prompts the agent runs on a timer | `cade cron list / add / remove / enable / disable / run-now` |
-| **Channels** | JSONL-backed inter-session pub/sub | `cade channels list / send / recv / clear` |
 
-See [`docs/channels.md`](./docs/channels.md) for the channels protocol; hooks and tasks/cron details live in their respective `cade <subcommand> --help` outputs.
+Cross-agent messaging is set up with `cade link` / `cade channel` (an agent can only
+reach another it's been linked to). Hooks and tasks/cron details live in their
+respective `cade <subcommand> --help` outputs.
 
 ---
 
@@ -171,9 +188,9 @@ Cade works with any OpenAI-compatible endpoint. Three ways to wire one in (CLI f
 # CLI flag for Ollama. Use a model from `ollama list`.
 cade chat -L --endpoint http://localhost:11434 --model qwen3:4b
 
-# Environment for one-shot mode
-cade context create local-qwen --model qwen3:4b
-cade context use local-qwen
+# Or create a dedicated agent for the local model
+cade new local-qwen --model qwen3:4b
+cade use local-qwen
 export OLLAMA_BASE_URL="http://localhost:11434"
 CADE_LOCAL_ONLY=1 cade -m "hello from a local model"
 
@@ -193,12 +210,11 @@ All state lives under `~/.cade/`. Override the location with `CADE_HOME=/path/to
 
 | File | Purpose |
 |---|---|
-| `config/cade.toml` | Model / provider / UI settings |
-| `data/contexts/<context>/history.db` | Thread + message storage |
+| `config/cade.toml` | Global settings + `[agent_defaults]` (model, provider, UI, …) |
+| `config/agents/<name>.toml` | Per-agent config (model, prompt, `[[mcp]]`, `[serve]`, hooks) |
+| `data/agents/<id>/` | Per-agent state: `threads/ memory/ tasks/ inbox/` |
+| `run/agents/<name>/` | Per-agent daemon runtime: `serve.pid`, mcp-services, cron lease |
 | `logs/cade.log` | Rotating log file |
-| `config/contexts/<context>.toml [[mcp]]` | User MCP server registry |
-| `config/contexts/<context>.toml [serve]` | `cade serve` daemon config |
-| `personas/`, `memory/`, `tasks/`, `channels/` | Persona-scoped state |
 
 Full layout, env vars, and TOML schema in [`docs/configuration.md`](./docs/configuration.md).
 

{cade_cli-0.13.4 → cade_cli-0.14.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "cade-cli"
-version = "0.13.4"
+version = "0.14.0"
 description = "Cade - The CLI Agent from Arcade.dev"
 readme = "README.md"
 requires-python = ">=3.11"

{cade_cli-0.13.4 → cade_cli-0.14.0}/src/cade_mcp_local/server.py

@@ -19,6 +19,9 @@ patch_frozen_return_detection()
 
 from cade_mcp_local.resources import register_resources  # noqa: E402
 from cade_mcp_local.tools import (  # noqa: E402
+    agent_list_tool,
+    agent_send_tool,
+    agent_spawn_tool,
     ask_user_question_tool,
     bash_tool,
     edit_tool,
@@ -54,6 +57,10 @@ is running. Use these tools when you need to:
   TaskUpdate, TaskDelete) — the host registers them as Local_* names
 - PushNotification to inject a structured notice into the agent context at the next
   tool boundary
+- AgentSend / AgentList to coordinate with other agents the user has linked to this
+  one (ACL-gated; delivered to the target agent's inbox)
+- AgentSpawn to delegate a self-contained sub-task to a fresh child agent and get
+  back just its result (requires this agent to allow spawning)
 - AskUserQuestion when you need structured clarification from the user (host:
   Local_AskUserQuestion)
 - WebFetch to read an external URL (Firecrawl via Arcade by default, httpx
@@ -112,6 +119,12 @@ def create_app() -> MCPApp:
     # Notifications (inject XML into context at next tool boundary)
     app.add_tool(push_notification_tool)
 
+    # Cross-agent coordination (ACL-gated; delivers to another agent's inbox)
+    app.add_tool(agent_send_tool)
+    app.add_tool(agent_list_tool)
+    # Spawn an ephemeral child task-agent (opt-in via the agent's allow_spawn)
+    app.add_tool(agent_spawn_tool)
+
     # Structured clarification question for the user
     app.add_tool(ask_user_question_tool)
 

{cade_cli-0.13.4 → cade_cli-0.14.0}/src/cade_mcp_local/tools/__init__.py

@@ -5,6 +5,7 @@ is running. Use these tools when you need to read, write, search, or execute
 commands on the same machine as the running cade process.
 """
 
+from cade_mcp_local.tools.agents import agent_list_tool, agent_send_tool, agent_spawn_tool
 from cade_mcp_local.tools.context import search_recent_context_tool
 from cade_mcp_local.tools.filesystem import (
     edit_tool,
@@ -62,6 +63,10 @@ __all__ = [
     "task_delete_tool",
     # Notifications
     "push_notification_tool",
+    # Cross-agent coordination
+    "agent_send_tool",
+    "agent_list_tool",
+    "agent_spawn_tool",
     # Interactive
     "ask_user_question_tool",
     # Web

cade_cli-0.14.0/src/cade_mcp_local/tools/agents.py

@@ -0,0 +1,151 @@
+"""MCP tools for cross-agent coordination.
+
+Lets the model address other agents the user has explicitly linked. Delivery
+rides the durable per-agent inbox (so the message survives until the target
+agent's next turn), and the user-provisioned channel ACL gates every send:
+two top-level agents can talk only if a user ran ``cade link`` / added them to
+a shared channel. The default is isolation.
+"""
+
+import logging
+from typing import Annotated, Any
+
+from arcade_mcp_server import Context, tool
+from typing_extensions import TypedDict
+
+from cade_mcp_local._metadata import CREATE_META, READ_META
+from cade_mcp_local.errors import InvalidInputError
+from cadecoder.core import agent as agent_mod
+from cadecoder.core import comms as comms_mod
+from cadecoder.core.inbox import inbox_for_path
+from cadecoder.core.paths import get_paths
+
+log = logging.getLogger(__name__)
+
+
+class AgentSendOutput(TypedDict):
+    delivered: bool
+    target: str
+    target_id: str
+
+
+class AgentSpawnOutput(TypedDict):
+    agent: str
+    id: str
+    result: str
+
+
+class AgentListOutput(TypedDict):
+    agents: list[dict[str, Any]]
+    count: int
+
+
+def _agent_from_request_meta() -> str | None:
+    """Per-request acting agent from MCP ``_meta['cade/agent']``, if the host set it.
+
+    The cade host stamps this on every built-in ``tools/call`` so the agent
+    resolves per-request rather than from this subprocess's env/state. Returns
+    ``None`` when running outside a request context (tests, one-shot CLI).
+    """
+    try:
+        from arcade_mcp_server.request_context import get_request_meta
+    except Exception:
+        return None
+    meta = get_request_meta()
+    if meta is None:
+        return None
+    return meta.__dict__.get("cade/agent")
+
+
+def _current_agent_name() -> str:
+    name = _agent_from_request_meta() or agent_mod.get_active_agent_name()
+    if not name:
+        raise InvalidInputError("agent", "no current agent is set")
+    return name
+
+
+@tool(
+    name="AgentSend",
+    desc=(
+        "Send a message to another agent's inbox. Only works for agents a user "
+        "has linked to this one (via `cade link` or a shared channel); otherwise "
+        "the send is refused. The message is delivered at the target agent's next "
+        "turn. Use AgentList first to see who you can reach."
+    ),
+    metadata=CREATE_META,
+)
+def agent_send_tool(
+    context: Context,
+    target: Annotated[str, "Target agent name, id, or unique id prefix."],
+    message: Annotated[str, "Message to deliver to the target agent."],
+) -> Annotated[AgentSendOutput, "Delivery result."]:
+    if not message.strip():
+        raise InvalidInputError("message", "cannot be empty")
+    sender = _current_agent_name()
+    target_agent = agent_mod.resolve(target)
+    if target_agent is None:
+        raise InvalidInputError("target", f"agent '{target}' not found")
+    if not comms_mod.can_communicate(sender, target_agent.name):
+        raise InvalidInputError(
+            "target",
+            f"not linked with '{target_agent.name}'. A user must run: "
+            f"cade link {sender} {target_agent.name}",
+        )
+    inbox = inbox_for_path(get_paths().agent_data(target_agent.id))
+    inbox.enqueue_notification(message, source=f"agent:{sender}")
+    return {"delivered": True, "target": target_agent.name, "target_id": target_agent.id}
+
+
+@tool(
+    name="AgentSpawn",
+    desc=(
+        "Delegate a focused sub-task to a fresh child agent that runs in its own "
+        "process and context, returning just the result (keeps your context "
+        "clean). Only works if this agent was created with --allow-spawn. Blocks "
+        "until the child finishes."
+    ),
+    metadata=CREATE_META,
+)
+async def agent_spawn_tool(
+    context: Context,
+    task: Annotated[str, "The self-contained sub-task for the child agent."],
+) -> Annotated[AgentSpawnOutput, "The child agent and its result."]:
+    from cadecoder.core.spawn import SpawnError, spawn_task_agent
+
+    if not task.strip():
+        raise InvalidInputError("task", "cannot be empty")
+    # Resolve the parent from the per-request agent identity so the child is
+    # attributed to the calling agent, not this subprocess's env/state.
+    parent_name = _agent_from_request_meta()
+    parent = agent_mod.resolve(parent_name) if parent_name else None
+    try:
+        return await spawn_task_agent(task, parent=parent)  # type: ignore[return-value]
+    except SpawnError as e:
+        raise InvalidInputError("task", str(e)) from e
+
+
+@tool(
+    name="AgentList",
+    desc=(
+        "List the agents you can message — i.e. agents a user has linked to this "
+        "one through a shared channel. Returns an empty list when you are isolated."
+    ),
+    metadata=READ_META,
+)
+def agent_list_tool(
+    context: Context,
+) -> Annotated[AgentListOutput, "Reachable agents."]:
+    sender = _current_agent_name()
+    reachable: list[dict[str, Any]] = []
+    for candidate in agent_mod.list_agents():
+        if candidate.name == sender:
+            continue
+        if comms_mod.can_communicate(sender, candidate.name):
+            reachable.append(
+                {
+                    "name": candidate.name,
+                    "id": candidate.id,
+                    "description": candidate.description,
+                }
+            )
+    return {"agents": reachable, "count": len(reachable)}

{cade_cli-0.13.4 → cade_cli-0.14.0}/src/cade_mcp_local/tools/notifications.py

@@ -9,7 +9,7 @@ from typing_extensions import TypedDict
 
 from cade_mcp_local._metadata import CREATE_META
 from cade_mcp_local.errors import InvalidInputError
-from cadecoder.tasks import NotificationQueue
+from cadecoder.core.inbox import current_inbox
 
 log = logging.getLogger(__name__)
 
@@ -45,5 +45,5 @@ def push_notification_tool(
         raise InvalidInputError("message", "cannot be empty")
     level_value = level.value if isinstance(level, NotificationLevel) else str(level)
     xml = f"<notification>\n  <level>{level_value}</level>\n  <message>{message}</message>\n</notification>"
-    NotificationQueue.get().enqueue(xml)
+    current_inbox().enqueue(xml)
     return {"queued": True, "level": level_value}