feat: fix aidsdk step taking, demo updates

apps/sandbox-agent/src/agent.ts

@@ -285,7 +285,6 @@ export default defineAgent({
       const memoryExecutor = new AiSdkExecutor<SandboxAttachmentData>({
         model: aiSdkOpenAI(DISCRETE_MODEL),
         maxOutputTokens: 1024,
-        maxSteps: 3,
         providerOptions: OPENAI_NO_REASONING_OPTIONS,
       });
       const isMemoryRequest = (request: { inference: { tools: Array<{ name: string }> } }) =>

apps/sandbox/convex/sessionActions.ts

@@ -67,7 +67,6 @@ export const sendMessage = action({
     const memoryExecutor = new AiSdkExecutor<SandboxAttachmentData>({
       model: aiSdkOpenAI(DISCRETE_MODEL),
       maxOutputTokens: 1024,
-      maxSteps: 3,
     });
     const isMemoryRequest = (request: { inference: { tools: Array<{ name: string }> } }) =>
       request.inference.tools.some((tool) => tool.name === "saveMemories");

apps/sandbox/convex/sessions.ts

@@ -273,7 +273,6 @@ function createObservabilityExecutor(): Executor<SandboxAttachmentData> {
   const memoryExecutor = new AiSdkExecutor<SandboxAttachmentData>({
     model: modelRef(DISCRETE_MODEL),
     maxOutputTokens: 1024,
-    maxSteps: 3,
   });
   const isMemoryRequest = (request: { inference: { tools: Array<{ name: string }> } }) =>
     request.inference.tools.some((tool) => tool.name === "saveMemories");

apps/site/README.md

@@ -10,9 +10,12 @@ Create `apps/site/.env.local` from `.env.local.example`, then start Convex:
 
 ```sh
 cd apps/site
-npx convex dev
+bun run convex
 ```
 
+This regenerates the read-only repository snapshot used by the demo agent
+before starting Convex.
+
 Set the model key in the Convex deployment environment, not in Vite:
 
 ```sh

apps/site/card-copy.md

@@ -0,0 +1,44 @@
+# Explainer card copy
+
+Proposed subtext for each category card under the hero. Current copy is Latin
+lorem; targets are one short line (~50–70 chars) to match the existing layout.
+
+Logos are citations, not labels. Two kinds in play:
+
+- **Wikipedia mark** → the concept's canonical article.
+- **Doc icon** (projector rainbow theme) → the concept's doc in the projector
+  GitHub repo. Both docs are stubbed at `packages/projector/docs/`.
+
+## framework
+
+| Card | Subtext | Logo → links to |
+| --- | --- | --- |
+| typescript | Fully typed end to end, frontend and backend. | — |
+| component based | Local reasoning for agents. Compose, nest, and reuse them. | — |
+| context projection | Model the full resolution of reality, then choose how it projects to the llm. | — |
+| state complete | Frame log + app state + configuration. All replayable, forkable, testable state. | Doc icon (rainbow) → [state-complete.md](https://github.com/anteprojector/projector/blob/main/packages/projector/docs/state-complete.md) |
+
+## evolutionary
+
+| Card | Subtext | Logo → links to |
+| --- | --- | --- |
+| self assembly | The agent can modify itself, and the app adapts to the user. | — |
+| evolutionary search | A new form of software development, inspired by evolution. | Wikipedia → [Evolutionary algorithm](https://en.wikipedia.org/wiki/Evolutionary_algorithm) |
+
+## apps
+
+| Card | Subtext | Logo → links to |
+| --- | --- | --- |
+| multi-actor | All agentic applications are multiplayer: the user and the llm. | — |
+| client server unification | Ship agentic features atomically across the client and server. | — |
+| endogenous state | Application state is more powerful when the agent owns it. | — |
+| generative ui | Made useful because endogenous state is the hard part. | — |
+
+## bonus cards (no category)
+
+| Card | Subtext | Logo → links to |
+| --- | --- | --- |
+| composable dimensions | Separate the base agent configuration from its thread configuration. Project into one at runtime. | — |
+| state concurrency | Allow the llm and the user to share state. | — |
+| durable frame log | Every step recorded, including state updates, work scheduling, and tool calls. | Doc icon (rainbow) → [durable-frame-log.md](https://github.com/anteprojector/projector/blob/main/packages/projector/docs/durable-frame-log.md) |
+| coordinated subagents | Agents are systems and need proper coordination primitives. | — |

apps/site/convex.json

@@ -1,5 +1,5 @@
 {
   "node": {
-    "externalPackages": []
+    "externalPackages": ["just-bash"]
   }
 }

apps/site/convex/_generated/api.d.ts

@@ -21,8 +21,12 @@ import type * as dev_sessions from "../dev/sessions.js";
 import type * as executorConfig from "../executorConfig.js";
 import type * as frameHistory from "../frameHistory.js";
 import type * as http from "../http.js";
+import type * as inbox from "../inbox.js";
 import type * as messageActor from "../messageActor.js";
 import type * as messages from "../messages.js";
+import type * as repoBash from "../repoBash.js";
+import type * as runnerShared from "../runnerShared.js";
+import type * as sessionEphemera from "../sessionEphemera.js";
 import type * as sessionParticipants from "../sessionParticipants.js";
 import type * as sessions from "../sessions.js";
 import type * as topics from "../topics.js";
@@ -47,8 +51,12 @@ declare const fullApi: ApiFromModules<{
   executorConfig: typeof executorConfig;
   frameHistory: typeof frameHistory;
   http: typeof http;
+  inbox: typeof inbox;
   messageActor: typeof messageActor;
   messages: typeof messages;
+  repoBash: typeof repoBash;
+  runnerShared: typeof runnerShared;
+  sessionEphemera: typeof sessionEphemera;
   sessionParticipants: typeof sessionParticipants;
   sessions: typeof sessions;
   topics: typeof topics;