@columenlabs/lumen-ai 0.1.217 → 0.1.219

package/README.md

@@ -241,8 +241,12 @@ Drafts and writes are cards in the chat's own storage (`lumen_cards` in the agen
 
 **The panel.** `LumenAssistant` reads a Think chat with no change. `ask(input, sink)` runs one answer and sends the panel's own events (`AssistantEvent`: the chat, the text, each tool line, cards, hand-off sections, advice, the use) to `sink`. The app's `POST /chat` returns `askResponse((sink) => chat.ask(input, sink), { signal, stop: () => chat.stop() })`, the same server-sent events `assistantRoutes` sends. `view()` gives a stored chat as the panel's items, and `usage()` gives the chat's use so far. A card's id is its tool call's, so the panel's Save, Dismiss and Undo go to `saveCard`, `dismissCard` and `undoCard`. The app keeps the list of chats, and each chat is its own agent, such as a facet of the person's store. `chunkEvents` and `itemsOf` are the mapping, for an app that streams some other way.
 
+**Steps.** One answer makes at most `plan.maxSteps` model calls (`MAX_STEPS`, 8). The last one runs with no tools and `LAST_STEP_NOTE`, so an answer always ends in words: what it found, and what is left when the task did not fit.
+
 **Steer and queue.** While an answer runs, the panel's box stays open: Enter steers the message into the answer (`steer(message)` on the agent; the app routes `POST /chats/:id/steer` `{ message }` to it and answers `{ steered }`), Alt+Enter queues it for after. A steer joins at the answer's next step, after the tool it is on, with `STEER_NOTE` before it for the model; the panel shows it there marked Steered (event `steered`), and `view()` keeps it in place (`SteerNote`, under the message the answer is to). A steer the answer ends before reading comes back as `unsteered` and goes first in the queue. `plan.steer: false` turns it off for a model; an app without the route (404) shows Queue only. Queued messages each go as their own message when the answer before them ends; each has Steer now, Edit and Remove.
 
+**Retry.** A turn that ended in an error shows a Retry icon (⟳) beside Copy, always visible, while no answer runs and someone can pay (`status.wait` unset; hidden otherwise, never disabled). It sends the turn's words, `@` parts, `#` tier and images again as a new message (`chat.retry(at, user, extra)` in `useAssistantChat`), marked Retry above the bubble; the old error turns grey (`retried` on the error item) and stops offering Retry. Nothing changes on the server: the retried message is an ordinary message.
+
 ```ts
 class Chat extends LumenAgent<Env> {
   #settle?: (end: AgentTurnEnd) => Promise<void>;

package/dist/agent/agent.d.ts

@@ -74,7 +74,7 @@ export interface AgentTurnPlan {
     left?: number;
     /** The person lets the assistant move their screen: the `screen` tool joins when the browser sent its map. */
     screen?: boolean;
-    /** Most model calls in one answer; 8. */
+    /** Most model calls in one answer (`MAX_STEPS`, 8); the last one answers in words, with no tools. */
     maxSteps?: number;
     /**
      * A message sent while the answer runs joins it at its next step (Steer); on unless `false`,
@@ -99,6 +99,10 @@ export interface AgentTurnEnd {
     capped: boolean;
 }
 export declare const CAP_NOTE = "The spend cap for this month is reached: answer now from what you have, in a sentence or two. You can call no more tools.";
+/** The last call of an answer: no tools, so the answer never ends on a tool's result with nothing said. */
+export declare const LAST_STEP_NOTE = "This is the answer's last call and you can call no more tools: answer now from what you found. If the task is not finished, say what is done, what is left and that you can carry on when asked.";
+/** Most model calls in one answer when the plan names none. */
+export declare const MAX_STEPS = 8;
 /** Put before a message the person steered into a running answer, for the model only. */
 export declare const STEER_NOTE = "The person sent this while you were working. Take it into account from here on; it does not cancel what you were asked before unless it says so.";
 /** A card in the chat: a draft the person saves or dismisses, or a write made at once (Undo). */

package/dist/agent/agent.js

@@ -24,6 +24,10 @@ import { chunkEvents, itemsOf } from './stream.js';
 import { openAiTools, } from './tools.js';
 import { UsageMeter } from './usage.js';
 export const CAP_NOTE = 'The spend cap for this month is reached: answer now from what you have, in a sentence or two. You can call no more tools.';
+/** The last call of an answer: no tools, so the answer never ends on a tool's result with nothing said. */
+export const LAST_STEP_NOTE = "This is the answer's last call and you can call no more tools: answer now from what you found. If the task is not finished, say what is done, what is left and that you can carry on when asked.";
+/** Most model calls in one answer when the plan names none. */
+export const MAX_STEPS = 8;
 /** Put before a message the person steered into a running answer, for the model only. */
 export const STEER_NOTE = 'The person sent this while you were working. Take it into account from here on; it does not cancel what you were asked before unless it says so.';
 /** Past this many tokens in the chat Think summarizes its older part. */
@@ -596,7 +600,7 @@ export class LumenAgent extends Think {
             tools,
             ...(lazy ? { activeTools: lazy.active() } : {}),
             ...(plan.first ? { toolChoice: { type: 'tool', toolName: plan.first } } : {}),
-            maxSteps: plan.maxSteps ?? 8,
+            maxSteps: plan.maxSteps ?? MAX_STEPS,
             ...turnSettings(plan.model, {
                 effort: plan.effort ?? 'default',
                 ...(plan.out ? { out: plan.out } : {}),
@@ -620,6 +624,17 @@ export class LumenAgent extends Think {
             };
             return capped;
         }
+        // The last call answers in words: past it the loop stops, so a tool call there would end the
+        // answer on a tool's result with nothing said.
+        if (ctx.stepNumber >= (t.plan.maxSteps ?? MAX_STEPS) - 1) {
+            const last = {
+                ...messages,
+                activeTools: [],
+                toolChoice: 'none',
+                instructions: `${t.system}\n\n${LAST_STEP_NOTE}`,
+            };
+            return last;
+        }
         // A forced first call is for the first step only; tools a search found join each step.
         const auto = t.plan.first && ctx.stepNumber > 0;
         if (!auto && !t.lazy && !steered)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@columenlabs/lumen-ai",
-  "version": "0.1.217",
+  "version": "0.1.219",
   "description": "Lumen's assistant and agent parts: a chat assistant's server (Claude tiers, tools over MCP, the web, hand off, spend caps), serving an MCP server from a Worker, read-only tool sets and metered tool calls.",
   "license": "MIT",
   "type": "module",
@@ -52,8 +52,8 @@
     "@ai-sdk/openai": "4.0.108",
     "@anthropic-ai/sdk": "0.133.0",
     "@cloudflare/think": "0.23.0",
-    "@columenlabs/lumen-ops": "0.1.217",
-    "@columenlabs/lumen-ui-types": "0.1.217",
+    "@columenlabs/lumen-ops": "0.1.219",
+    "@columenlabs/lumen-ui-types": "0.1.219",
     "agents": "0.32.0",
     "ai": "7.0.162",
     "just-bash": "3.6.0",