v0.7.0: Add multi-agent orchestration with CLI and API providers

Each role (PI, Trainee, Reviewer) can now run as a separate agent with
its own context window. Opt-in via `orchestration: multi` in config.yaml.

CLI providers (claude-cli, codex-cli, cursor-cli) are the primary path:
zero API key config, uses existing subscriptions. API providers
(anthropic, openai) are the fallback for per-token billing.

Graceful fallback: if a CLI binary is missing or any error occurs,
falls through to single-agent mode for that turn.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: albert-ying

README.md

@@ -90,6 +90,7 @@ When work is ready, you review it. Accept, request revisions, or reject. The loo
 ## Key capabilities
 
 - **Zero additional cost**: runs on your existing coding agent subscription. No separate API keys, no usage-based billing, no new accounts.
+- **Multi-agent orchestration**: opt-in mode where each role (PI, Trainee, Reviewer) runs as a separate agent with its own context window. Uses your existing CLI subscriptions (Claude Code, Codex CLI, Cursor) -- no API keys needed. Falls back to single-agent on any failure.
 - **Skill containers**: configure characters with any combination of SKILL.md files you already have. A PI with `scanpy + scientific-writing + statistical-analysis` skills behaves differently from a Tech Lead with `react + typescript + code-review` skills.
 - **24-hour sessions**: the loop runs indefinitely. No timeout, no context loss. Sessions persist across disconnects with `autolab_resume`.
 - **Fully configurable**: YAML character profiles control personality, expertise, goals, and available tools. Swap them in seconds.
@@ -132,6 +133,40 @@ personality:
   - "Rigorous: demands statistical reproducibility"
 ```
 
+## Multi-agent mode
+
+By default, a single AI agent plays all roles in sequence. Multi-agent mode spawns a dedicated agent for each role, giving each a fresh context window. Enable it in `.autolab/config.yaml`:
+
+```yaml
+orchestration: multi
+
+agents:
+  pi:
+    provider: claude-cli   # Uses your Claude Code subscription
+    model: opus
+  trainee:
+    provider: claude-cli
+    model: sonnet
+```
+
+Supported providers:
+
+| Provider | Cost | Requires |
+|----------|------|----------|
+| `claude-cli` | Free (Pro/Max sub) | Claude Code CLI installed |
+| `codex-cli` | Free (Plus/Pro sub) | Codex CLI installed |
+| `cursor-cli` | Free (Pro/Biz sub) | Cursor Agent CLI installed |
+| `anthropic` | Pay-per-token | `ANTHROPIC_API_KEY` env var |
+| `openai` | Pay-per-token | `OPENAI_API_KEY` env var |
+
+CLI providers are the primary path -- they need zero API key configuration. If the CLI binary is not found or multi-agent fails for any reason, it falls back to single-agent mode automatically.
+
+For API providers, install the optional dependencies:
+
+```bash
+pip install 'autonomous-lab[multi-agent]'
+```
+
 ## Remote / SSH environments
 
 The monitoring web UI binds to `127.0.0.1` by default (local only). On a remote server, SSH session, or container, the UI will attempt to auto-detect and bind to `0.0.0.0` instead. If auto-detection doesn't match your setup, use one of the methods below.

docs/index.html

@@ -20,6 +20,7 @@
         </div>
         <div class="nav-links">
             <a href="#hero" class="nav-link active">Home</a>
+            <a href="#multi-agent" class="nav-link">Multi-Agent</a>
             <a href="#tutorial" class="nav-link">Tutorial</a>
             <a href="#characters" class="nav-link">Characters</a>
             <a href="#marketplace" class="nav-link">Marketplace</a>
@@ -99,6 +100,49 @@ <h1 class="hero-title">Autonomous Lab</h1>
                 <span class="badge">&#x1F9EC; Skill Containers</span>
                 <span class="badge">&#x23F1;&#xFE0F; 24-Hour Sessions</span>
                 <span class="badge">&#x2699;&#xFE0F; Fully Configurable</span>
+                <span class="badge">&#x1F916; Multi-Agent</span>
+            </div>
+        </div>
+    </section>
+
+    <!-- ========== MULTI-AGENT SECTION ========== -->
+    <section id="multi-agent" class="section">
+        <div class="section-header">
+            <h2 class="section-title">&#x1F916; Multi-Agent Orchestration</h2>
+            <p class="section-subtitle">Each role gets its own context window &mdash; using the CLIs you already have</p>
+        </div>
+
+        <div class="gs-grid">
+            <div class="gs-card">
+                <div class="gs-card-num">&#x1F50C;</div>
+                <h3>Zero Extra Cost</h3>
+                <p>Multi-agent mode spawns separate CLI agents (<code>claude</code>, <code>codex</code>, <code>cursor-agent</code>) for each role. It uses the subscriptions you already pay for &mdash; Claude Pro, ChatGPT Plus, Cursor Pro. No API keys needed.</p>
+            </div>
+
+            <div class="gs-card">
+                <div class="gs-card-num">&#x1F9E0;</div>
+                <h3>Fresh Context Per Role</h3>
+                <p>In single-agent mode, one context window fills with mixed PI and Trainee concerns. Multi-agent gives each role a dedicated context, so the PI thinks like a PI and the Trainee codes like a Trainee.</p>
+            </div>
+
+            <div class="gs-card">
+                <div class="gs-card-num">&#x2699;</div>
+                <h3>One Line to Enable</h3>
+                <p>Add <code>orchestration: multi</code> to <code>.autolab/config.yaml</code>. That&rsquo;s it. If the CLI isn&rsquo;t installed, it falls back to single-agent automatically. You can also mix providers &mdash; e.g. Claude as PI, Codex as Trainee.</p>
+                <pre class="gs-code"><code><span class="yaml-key">orchestration:</span> <span class="yaml-val">multi</span>
+<span class="yaml-key">agents:</span>
+  <span class="yaml-key">pi:</span>
+    <span class="yaml-key">provider:</span> <span class="yaml-val">claude-cli</span>
+    <span class="yaml-key">model:</span> <span class="yaml-val">opus</span>
+  <span class="yaml-key">trainee:</span>
+    <span class="yaml-key">provider:</span> <span class="yaml-val">codex-cli</span>
+    <span class="yaml-key">model:</span> <span class="yaml-val">o3</span></code></pre>
+            </div>
+
+            <div class="gs-card">
+                <div class="gs-card-num">&#x1F6E1;</div>
+                <h3>Graceful Fallback</h3>
+                <p>If a CLI binary is missing, an API key isn&rsquo;t set, or anything goes wrong &mdash; the system automatically falls back to single-agent mode for that turn. No configuration changes, no broken sessions. It just works.</p>
             </div>
         </div>
     </section>

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "autonomous-lab"
-version = "0.6.0"
+version = "0.7.0"
 description = "MCP server that turns any Senior-Junior workflow into an autonomous loop with a human decision maker. Seamless integration with Cursor, Claude Code, Codex, and any MCP client."
 readme = "README.md"
 requires-python = ">=3.11"
@@ -39,6 +39,8 @@ Issues = "https://github.com/albert-ying/autonomous-lab/issues"
 
 [project.optional-dependencies]
 biotools = ["biomni"]
+multi-agent = ["anthropic>=0.40.0", "openai>=1.50.0"]
+all = ["biomni", "anthropic>=0.40.0", "openai>=1.50.0"]
 
 [project.scripts]
 autonomous-lab = "autonomous_lab.__main__:main"

src/autonomous_lab/__init__.py

@@ -6,7 +6,7 @@
 Fork of mcp-feedback-enhanced with virtual lab extensions.
 """
 
-__version__ = "0.6.0"
+__version__ = "0.7.0"
 __author__ = "Albert Ying"
 
 from .server import main as run_server