Support rcrewai 0.10 and fix the docs and install template

.github/workflows/ci.yml

@@ -17,7 +17,7 @@ jobs:
     steps:
       - uses: actions/checkout@v4
 
-      # rcrewai now ships to rubygems, so the gemspec dep (`rcrewai ~> 0.8.0`)
+      # rcrewai now ships to rubygems, so the gemspec dep (`rcrewai ~> 0.10.0`)
       # resolves from there. The sibling-checkout workaround this job used while
       # the gem was unpublished has been removed.
       - name: Set up Ruby

CHANGELOG.md

@@ -7,6 +7,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Changed
+- Requires `rcrewai ~> 0.10.0`. No migration is needed, and the suite passed against it unchanged. The engine does not expose 0.10's `CancelToken`, per-call `chat(model:)` or `cached_tokens`; see the README.
+
+### Fixed
+- **Docs and install template:** the README and docs site set `config.openai_api_key` inside `RcrewAI.configure`, which is the engine's configuration, so the example raised `NoMethodError`. They now use `RCrewAI.configure(validate: false)`. The generated initializer's commented example also named settings that don't exist (`groq_api_key`, `default_model`).
+- **Docs:** the README's `tools [ ... ]` example nested the array and broke the agent, and it didn't say that tools must be required explicitly.
+- **Docs:** corrected the `CrewExecutionJob` namespace, the rcrewai repo link and the gem's documentation URL, listed `POST /crews`, and brought the docs site up from rcrewai 0.7.
+
 ## [0.9.0] - 2026-09-10
 
 Tracks **rcrewai 0.9.0**, whose theme is concurrency: a turn's tool calls,

README.md

@@ -1,6 +1,6 @@
 # RcrewAI Rails
 
-Rails engine for integrating [RcrewAI](https://github.com/gkosmo/rcrewai-rails) into your Rails applications. Provides ActiveRecord persistence, background job integration, generators, and a web UI for managing AI crews and agents.
+Rails engine for integrating [rcrewai](https://github.com/gkosmo/rcrewAI) into your Rails applications. Provides ActiveRecord persistence, background job integration, generators, and a web UI for managing AI crews and agents.
 
 ## Features
 
@@ -10,7 +10,7 @@ Rails engine for integrating [RcrewAI](https://github.com/gkosmo/rcrewai-rails)
 - **Web UI**: Monitor and manage crews through a built-in interface
 - **Rails-Specific Tools**: Pre-built tools for ActiveRecord, ActionMailer, Rails cache, and more
 - **Configuration**: Flexible configuration through Rails initializers
-- **Full rcrewai 0.9 feature coverage** (see [rcrewai 0.9](#rcrewai-09-capabilities), [0.8](#rcrewai-08-capabilities) and [0.7](#rcrewai-07-capabilities) capabilities):
+- **Runs on rcrewai 0.10, with features through 0.9 exposed on the models** (see [rcrewai 0.10](#rcrewai-010), [0.9](#rcrewai-09-capabilities), [0.8](#rcrewai-08-capabilities) and [0.7](#rcrewai-07-capabilities)):
   - Agent config: reasoning, per-agent LLM, rate limiting, context-window trimming, cognitive memory
   - Task output: structured output schemas, guardrails, file output, multimodal attachments
   - Crew: `before_kickoff`/`after_kickoff` hooks, planning, the `consensual` process, batch execution
@@ -171,8 +171,12 @@ RcrewAI::Rails.configure do |config|
   config.log_level = :info
 end
 
-# Configure the base RcrewAI gem
-RcrewAI.configure do |config|
+# Configure the base rcrewai gem. It reads provider keys such as OPENAI_API_KEY
+# and ANTHROPIC_API_KEY from the environment on its own, so this is only needed
+# for values set in code. Note the capital C: `RcrewAI.configure` is the engine's
+# configuration, which has no provider settings. `validate: false` stops a missing
+# key from failing boot in places that don't call an LLM, such as assets:precompile.
+RCrewAI.configure(validate: false) do |config|
   config.openai_api_key = ENV["OPENAI_API_KEY"]
   # Add other LLM provider keys as needed
 end
@@ -239,14 +243,20 @@ execution = crew.execute(topic: "AI in Healthcare")
 
 ### Using Rails-Specific Tools
 
+The tools are not loaded with the engine, so require the ones you use:
+
 ```ruby
+require "rcrewai/rails/tools/active_record_tool"
+require "rcrewai/rails/tools/rails_cache_tool"
+require "rcrewai/rails/tools/action_mailer_tool"
+
 class DataAnalystAgent
   include RcrewAI::Rails::AgentBuilder
 
   agent_role "Data Analyst"
   agent_goal "Analyze application data"
 
-  tools [
+  tools(
     RcrewAI::Rails::Tools::ActiveRecordTool.new(
       model_class: User,
       allowed_methods: [:count, :where, :pluck]
@@ -256,10 +266,13 @@ class DataAnalystAgent
       mailer_class: ReportMailer,
       allowed_methods: [:send_report]
     )
-  ]
+  )
 end
 ```
 
+Pass the tools as separate arguments, not as one array: `tools [a, b]` nests the
+array and the agent fails when it lists its tools.
+
 ### Monitoring Executions
 
 Access the web UI at `/rcrewai` to:
@@ -280,12 +293,35 @@ crew.execute_async(inputs)
 crew.execute_sync(inputs)
 
 # Custom job options
-CrewExecutionJob.set(wait: 5.minutes).perform_later(crew, inputs)
+RcrewAI::Rails::CrewExecutionJob.set(wait: 5.minutes).perform_later(crew, inputs)
 ```
 
-## rcrewai 0.9 capabilities
+## rcrewai 0.10
+
+This engine requires [rcrewai](https://github.com/gkosmo/rcrewAI) `~> 0.10.0`. The
+upgrade needs no migration. rcrewai 0.10 adds live streaming features to its
+Anthropic client, and most of them work at the client level, so the engine has no
+settings for them:
+
+| rcrewai 0.10 feature | In this engine |
+|---|---|
+| Keep-alive connection for streaming requests | Applies automatically to Anthropic agents that stream |
+| `config.base_url` for Anthropic | Set it in `RCrewAI.configure` |
+| `RCrewAI::CancelToken` / `chat(cancel:)` | Not exposed. `Execution#cancel!` only updates the record's status and does not stop a running stream |
+| `chat(model:)` per-call override | Not exposed. Use an agent's `llm_config` to choose its model |
+| `cached_tokens` on usage | Not recorded. The observation engine does not store it on spans or rollups |
 
-This engine tracks [rcrewai](https://github.com/gkosmo/rcrewAI) `~> 0.9`.
+To use cancellation or per-call models, call an rcrewai client directly from your
+own code.
+
+**Current Claude models:** these reject the `temperature` that rcrewai sends by
+default, so turn it off globally. A per-agent `llm_config` can't remove it.
+
+```ruby
+RCrewAI.configure(validate: false) { |config| config.temperature = nil }
+```
+
+## rcrewai 0.9 capabilities
 
 ### Concurrency
 
@@ -397,15 +433,16 @@ existing records are unaffected — set only what you need.
 | `reasoning` / `max_reasoning_attempts` | Run a planning/reasoning pass before answering |
 | `respect_context_window` | Trim history to fit the model's context window |
 | `parallel_tools` | Run a turn's tool calls concurrently (rcrewai 0.9; `nil` uses the gem default of on) |
-| `llm_config` (JSON) | Per-agent LLM override, e.g. `{ "provider": "anthropic", "model": "claude-sonnet-5" }` |
+| `llm_config` (JSON) | Per-agent LLM override, e.g. `{ "provider": "anthropic", "model": "claude-opus-5-5" }` (see the [`temperature` note](#rcrewai-010)) |
 | `memory_enabled` + `memory_scope` + `memory_short_term_limit` | Enable cognitive memory (see below) |
 
 ```ruby
 agent = crew.agents.create!(
   name: "researcher", role: "Researcher", goal: "Find facts",
   reasoning: true,
   max_rpm: 30,
-  llm_config: { provider: "anthropic", model: "claude-sonnet-5" },
+  # Current Claude models also need RCrewAI's temperature set to nil; see "rcrewai 0.10".
+  llm_config: { provider: "anthropic", model: "claude-opus-5-5" },
   memory_enabled: true, memory_scope: "research", memory_short_term_limit: 20
 )
 ```
@@ -500,6 +537,7 @@ The engine provides JSON API endpoints:
 ```
 GET    /rcrewai/api/v1/crews
 GET    /rcrewai/api/v1/crews/:id
+POST   /rcrewai/api/v1/crews
 POST   /rcrewai/api/v1/crews/:id/execute
 GET    /rcrewai/api/v1/executions
 GET    /rcrewai/api/v1/executions/:id

docs/api.html

@@ -186,6 +186,7 @@ <h2>JSON API endpoints</h2>
       <tbody>
         <tr><td><span class="verb get">GET</span></td><td class="method">/crews</td></tr>
         <tr><td><span class="verb get">GET</span></td><td class="method">/crews/:id</td></tr>
+        <tr><td><span class="verb post">POST</span></td><td class="method">/crews</td></tr>
         <tr><td><span class="verb post">POST</span></td><td class="method">/crews/:id/execute</td></tr>
         <tr><td><span class="verb get">GET</span></td><td class="method">/executions</td></tr>
         <tr><td><span class="verb get">GET</span></td><td class="method">/executions/:id</td></tr>
@@ -204,7 +205,7 @@ <h2>Running executions</h2>
 crew.execute_sync(inputs)    # run inline
 
 # Custom job options:
-CrewExecutionJob.set(wait: 5.minutes).perform_later(crew, inputs)</pre>
+RcrewAI::Rails::CrewExecutionJob.set(wait: 5.minutes).perform_later(crew, inputs)</pre>
   </div>
 </section>
 

docs/capabilities.html

@@ -156,7 +156,7 @@
 <section class="page-hero">
   <div class="wrap">
     <h1>Capabilities</h1>
-    <p>The engine tracks rcrewai ~> 0.7. Every capability is configured through columns on the persisted models and forwarded to the core objects at build time — all off/absent by default.</p>
+    <p>The engine requires rcrewai ~> 0.10.0. Every capability is configured through columns on the persisted models and forwarded to the core objects at build time — all off/absent by default.</p>
   </div>
 </section>
 
@@ -170,6 +170,7 @@ <h2>Agent configuration</h2>
         <tr><td><code>max_rpm</code></td><td>Rate-limit the agent's LLM calls (requests/minute).</td></tr>
         <tr><td><code>reasoning</code> / <code>max_reasoning_attempts</code></td><td>Run a reasoning pass before answering.</td></tr>
         <tr><td><code>respect_context_window</code></td><td>Trim history to fit the model's context window.</td></tr>
+        <tr><td><code>parallel_tools</code></td><td>Run a turn's tool calls concurrently (nil uses the gem default of on).</td></tr>
         <tr><td><code>llm_config</code> (JSON)</td><td>Per-agent LLM override.</td></tr>
         <tr><td><code>memory_enabled</code> + <code>memory_scope</code> + <code>memory_short_term_limit</code></td><td>Enable cognitive memory.</td></tr>
       </tbody>
@@ -178,7 +179,8 @@ <h2>Agent configuration</h2>
   name: "researcher", role: "Researcher", goal: "Find facts",
   reasoning: true,
   max_rpm: 30,
-  llm_config: { provider: "anthropic", model: "claude-sonnet-5" },
+  # Current Claude models also need RCrewAI's temperature set to nil.
+  llm_config: { provider: "anthropic", model: "claude-opus-5-5" },
   memory_enabled: true, memory_scope: "research", memory_short_term_limit: 20
 )</pre>
   </div>
@@ -253,6 +255,73 @@ <h2>Flows</h2>
   </div>
 </section>
 
+<section>
+  <div class="wrap">
+    <h2>Observation engine</h2>
+    <p>Every execution is traced as a tree of spans (crew, agent, task, LLM call, and tool call) with timings, token counts, and cost. See the trace at <code>/rcrewai/executions/:id/observation</code> and spend at <code>/rcrewai/observations/costs</code>.</p>
+    <pre>RcrewAI::Rails.configure do |config|
+  config.observation_enabled = true
+  config.observation_capture_prompts = :truncated  # :none | :truncated | :full
+  config.observation_retention_days = 30
+end</pre>
+    <p class="caption">Prune old spans with rake rcrewai:observation:prune. Token and cost data need an agent that runs through rcrewai's native tool runner; with a fallback such as Ollama without native tools, the cost dashboard is empty.</p>
+  </div>
+</section>
+
+<section>
+  <div class="wrap">
+    <h2>Checkpointing and resume</h2>
+    <p>Record each task's result so an interrupted run can resume without re-running and re-paying for completed tasks. Off by default; enable it globally or per crew. It needs the <code>rcrewai_checkpoints</code> table.</p>
+    <pre>RcrewAI::Rails.configure { |config| config.checkpoint_enabled = true }  # globally
+crew.update!(checkpoint_enabled: true)                                   # or per crew</pre>
+  </div>
+</section>
+
+<section>
+  <div class="wrap">
+    <h2>Concurrent tool calls</h2>
+    <p>rcrewai runs a turn's tool calls concurrently by default. Opt out on an agent whose tools aren't safe to run in parallel, or that must not fan out against a rate-limited API. Left <code>nil</code>, the gem's default applies.</p>
+    <pre>agent.update!(parallel_tools: false)</pre>
+  </div>
+</section>
+
+<section>
+  <div class="wrap">
+    <h2>LLM interceptors</h2>
+    <p>Hooks that run around every LLM request the engine's agents make. Return a replacement to modify the payload or result, or <code>nil</code> to leave it untouched. A hook that raises is reported and skipped.</p>
+    <pre>RcrewAI::Rails.configure do |config|
+  config.llm_before_request = lambda do |payload, context|
+    Rails.logger.info("[llm] -> #{context[:provider]}/#{context[:model]}")
+    payload
+  end
+end</pre>
+  </div>
+</section>
+
+<section>
+  <div class="wrap">
+    <h2>Providers</h2>
+    <p>Set as an agent's <code>llm_config</code> provider or as <code>config.default_llm_provider</code>: <code>openai</code>, <code>anthropic</code>, <code>google</code>, <code>azure</code>, <code>ollama</code>, and:</p>
+    <table>
+      <thead><tr><th>Provider</th><th>Notes</th></tr></thead>
+      <tbody>
+        <tr><td><code>openai_compatible</code></td><td>Any OpenAI-format endpoint. Requires RCrewAI's <code>base_url</code>.</td></tr>
+        <tr><td><code>bedrock</code></td><td>AWS Bedrock via the Converse API. Requires <code>aws_region</code>.</td></tr>
+        <tr><td><code>snowflake</code></td><td>Snowflake Cortex inference. Requires <code>snowflake_account</code>.</td></tr>
+        <tr><td><code>openai_responses</code></td><td>OpenAI's Responses API. Non-streaming only.</td></tr>
+      </tbody>
+    </table>
+    <p class="caption">Current Claude models reject the temperature rcrewai sends by default, so set RCrewAI's temperature to nil to use them.</p>
+  </div>
+</section>
+
+<section>
+  <div class="wrap">
+    <h2>rcrewai 0.10</h2>
+    <p>Keep-alive streaming connections apply automatically to Anthropic agents that stream. <code>CancelToken</code>, per-call <code>chat(model:)</code>, and <code>cached_tokens</code> are not exposed by the engine: <code>Execution#cancel!</code> only updates the record's status and does not stop a running stream. Call an rcrewai client directly from your own code to use them.</p>
+  </div>
+</section>
+
 <footer>
   <div class="wrap">
     <p>

docs/getting-started.html

@@ -198,8 +198,9 @@ <h2>Configuration</h2>
   config.default_memory_store = RCrewAI::Memory::SqliteStore.new(path: "db/rcrewai_memory.sqlite3")
 end
 
-# Configure the base RcrewAI gem
-RcrewAI.configure do |config|
+# Configure the base rcrewai gem (note the capital C).
+# Keys in the environment are read automatically.
+RCrewAI.configure(validate: false) do |config|
   config.openai_api_key = ENV["OPENAI_API_KEY"]
 end</pre>
   </div>

docs/index.html

@@ -4,7 +4,7 @@
 <meta charset="utf-8">
 <meta name="viewport" content="width=device-width, initial-scale=1">
 <title>RcrewAI Rails — AI agent crews for Rails</title>
-<meta name="description" content="A Rails engine for orchestrating AI agents with ActiveRecord persistence, background jobs, generators, a web dashboard, and full rcrewai 0.7 feature coverage.">
+<meta name="description" content="A Rails engine for orchestrating AI agents with ActiveRecord persistence, background jobs, generators, a web dashboard, built on rcrewai 0.10.">
 <style>
   :root {
     --crimson: #c4221f;
@@ -156,7 +156,7 @@
 <section class="hero">
   <div class="wrap">
     <h1>Build AI agent crews,<br><span class="accent">the Rails way.</span></h1>
-    <p class="lead">A Rails engine for orchestrating AI agents — with ActiveRecord persistence, background-job execution, generators, a web dashboard, and full rcrewai 0.7 feature coverage.</p>
+    <p class="lead">A Rails engine for orchestrating AI agents — with ActiveRecord persistence, background-job execution, generators, a web dashboard, built on rcrewai 0.10.</p>
     <div class="install-line"><span class="prompt">gem </span>'rcrewai-rails'</div>
     <div class="cta">
       <a class="btn btn-primary" href="getting-started.html">Get Started</a>
@@ -193,7 +193,7 @@ <h3>Web Dashboard</h3>
       <div class="feature">
         <span class="ic">🤖</span>
         <h3>Multi-LLM Support</h3>
-        <p>OpenAI GPT, Anthropic Claude, Google Gemini, Azure OpenAI, and Ollama.</p>
+        <p>OpenAI, Anthropic Claude, Google Gemini, Azure OpenAI, Ollama, AWS Bedrock, Snowflake Cortex, and any OpenAI-compatible endpoint.</p>
       </div>
       <div class="feature">
         <span class="ic">🧰</span>
@@ -206,8 +206,8 @@ <h3>Rails-Specific Tools</h3>
 
 <section>
   <div class="wrap">
-    <h2>What's new <span class="badge">0.4 – 0.7</span></h2>
-    <p class="sub">Full parity with rcrewai 0.4–0.7, surfaced through ActiveRecord.</p>
+    <h2>What's new <span class="badge">0.4 – 0.10</span></h2>
+    <p class="sub">rcrewai features through 0.9, surfaced through ActiveRecord, on rcrewai 0.10.</p>
     <div class="whatsnew">
       <div class="wn-item">
         <h4>Agent intelligence</h4>
@@ -233,6 +233,22 @@ <h4>Flows</h4>
         <h4>Cognitive memory</h4>
         <p>Semantic recall with a pluggable embedder and SQLite-backed store.</p>
       </div>
+      <div class="wn-item">
+        <h4>Observation engine</h4>
+        <p>Span-level traces of every run, with timings, tokens, cost, and a live view.</p>
+      </div>
+      <div class="wn-item">
+        <h4>Checkpointing</h4>
+        <p>Resume an interrupted run without re-running the tasks it already finished.</p>
+      </div>
+      <div class="wn-item">
+        <h4>Interceptors and providers</h4>
+        <p>Hooks around every LLM request, plus Bedrock, Snowflake, and OpenAI-compatible endpoints.</p>
+      </div>
+      <div class="wn-item">
+        <h4>Concurrent tool calls</h4>
+        <p>A turn's tool calls run in parallel, with a per-agent opt-out.</p>
+      </div>
     </div>
   </div>
 </section>

lib/generators/rcrewai/rails/install/templates/rcrewai.rb

@@ -63,17 +63,25 @@
   # config.observation_retention_days = 30
 end
 
-# Configure RcrewAI base gem if needed
-RcrewAI.configure do |config|
-  # Set your API keys here or in environment variables
-  # config.openai_api_key = ENV["OPENAI_API_KEY"]
-  # config.anthropic_api_key = ENV["ANTHROPIC_API_KEY"]
-  # config.groq_api_key = ENV["GROQ_API_KEY"]
-
-  # Configure other RcrewAI settings
-  # config.default_model = "gpt-4"
-  # config.temperature = 0.7
+# Configure the base rcrewai gem if needed. Note the capital C: RCrewAI holds
+# provider settings, while RcrewAI.configure below is this engine's configuration.
+# validate: false stops a missing key from failing boot in places that don't call
+# an LLM, such as assets:precompile.
+RCrewAI.configure(validate: false) do |config|
+  # Keys are read from the environment (OPENAI_API_KEY, ANTHROPIC_API_KEY, ...)
+  # automatically, so set them here only if they come from somewhere else.
+  # config.openai_api_key = Rails.application.credentials.openai_api_key
+  # config.anthropic_api_key = Rails.application.credentials.anthropic_api_key
+
+  # config.openai_model = "gpt-4o"
+  # config.anthropic_model = "claude-opus-5-5"
 
+  # Current Claude models reject temperature, so leave it unset to use them.
+  # config.temperature = nil
+end
+
+# More engine settings. RcrewAI.configure is the same as RcrewAI::Rails.configure.
+RcrewAI.configure do |config|
   # --- Checkpointing (rcrewai 0.8+) -----------------------------------------
   # Records durable per-task state during a crew run so an interrupted run can
   # be resumed instead of re-executing (and re-paying for) completed tasks.