llmist 0.1.4 → 0.1.6

package/dist/cli.js

@@ -18,7 +18,9 @@ import {
   init_exceptions,
   init_logger,
   init_messages,
-  init_registry
+  init_model_shortcuts,
+  init_registry,
+  resolveModel
 } from "./chunk-J3NCIWMY.js";
 
 // src/cli/constants.ts
@@ -63,7 +65,7 @@ import { Command, InvalidArgumentError as InvalidArgumentError3 } from "commande
 // package.json
 var package_default = {
   name: "llmist",
-  version: "0.1.4",
+  version: "0.1.6",
   description: "Universal TypeScript LLM client with streaming-first agent framework. Works with any model - no structured outputs or native tool calling required. Implements its own flexible grammar for function calling.",
   type: "module",
   main: "dist/index.cjs",
@@ -169,11 +171,10 @@ init_builder();
 init_registry();
 init_constants();
 import { createInterface } from "node:readline/promises";
-import chalk3 from "chalk";
+import chalk2 from "chalk";
 import { InvalidArgumentError as InvalidArgumentError2 } from "commander";
 
 // src/cli/builtin-gadgets.ts
-import chalk from "chalk";
 import { z } from "zod";
 init_exceptions();
 var askUser = createGadget({
@@ -195,17 +196,17 @@ var tellUser = createGadget({
     type: z.enum(["info", "success", "warning", "error"]).default("info").describe("Message type: info, success, warning, or error")
   }),
   execute: ({ message, done, type }) => {
-    const formatters = {
-      info: (msg) => chalk.blue(`\u2139\uFE0F  ${msg}`),
-      success: (msg) => chalk.green(`\u2705 ${msg}`),
-      warning: (msg) => chalk.yellow(`\u26A0\uFE0F  ${msg}`),
-      error: (msg) => chalk.red(`\u274C ${msg}`)
+    const prefixes = {
+      info: "\u2139\uFE0F  ",
+      success: "\u2705 ",
+      warning: "\u26A0\uFE0F  ",
+      error: "\u274C "
     };
-    const formatted = formatters[type](message);
+    const plainResult = prefixes[type] + message;
     if (done) {
-      throw new BreakLoopException(formatted);
+      throw new BreakLoopException(plainResult);
     }
-    return formatted;
+    return plainResult;
   }
 });
 var builtinGadgets = [askUser, tellUser];
@@ -308,7 +309,7 @@ async function loadGadgets(specifiers, cwd, importer = (specifier) => import(spe
 
 // src/cli/utils.ts
 init_constants();
-import chalk2 from "chalk";
+import chalk from "chalk";
 import { InvalidArgumentError } from "commander";
 function createNumericParser({
   label,
@@ -366,9 +367,10 @@ function isInteractive(stream) {
 var SPINNER_FRAMES = ["\u280B", "\u2819", "\u2839", "\u2838", "\u283C", "\u2834", "\u2826", "\u2827", "\u2807", "\u280F"];
 var SPINNER_DELAY_MS = 500;
 var StreamProgress = class {
-  constructor(target, isTTY) {
+  constructor(target, isTTY, modelRegistry) {
     this.target = target;
     this.isTTY = isTTY;
+    this.modelRegistry = modelRegistry;
   }
   // Animation state
   frameIndex = 0;
@@ -389,6 +391,7 @@ var StreamProgress = class {
   // Cumulative stats (cumulative mode)
   totalStartTime = Date.now();
   totalTokens = 0;
+  totalCost = 0;
   iterations = 0;
   /**
    * Starts a new LLM call. Switches to streaming mode.
@@ -415,6 +418,20 @@ var StreamProgress = class {
     this.iterations++;
     if (usage) {
       this.totalTokens += usage.totalTokens;
+      if (this.modelRegistry && this.model) {
+        try {
+          const modelName = this.model.includes(":") ? this.model.split(":")[1] : this.model;
+          const cost = this.modelRegistry.estimateCost(
+            modelName,
+            usage.inputTokens,
+            usage.outputTokens
+          );
+          if (cost) {
+            this.totalCost += cost.totalCost;
+          }
+        } catch {
+        }
+      }
     }
     this.pause();
     this.mode = "cumulative";
@@ -478,33 +495,39 @@ var StreamProgress = class {
     const outTokens = this.callOutputTokensEstimated ? Math.round(this.callOutputChars / FALLBACK_CHARS_PER_TOKEN) : this.callOutputTokens;
     const parts = [];
     if (this.model) {
-      parts.push(chalk2.cyan(this.model));
+      parts.push(chalk.cyan(this.model));
     }
     if (this.callInputTokens > 0) {
       const prefix = this.callInputTokensEstimated ? "~" : "";
-      parts.push(chalk2.dim("out:") + chalk2.yellow(` ${prefix}${this.callInputTokens}`));
+      parts.push(chalk.dim("out:") + chalk.yellow(` ${prefix}${this.callInputTokens}`));
     }
     if (this.isStreaming || outTokens > 0) {
       const prefix = this.callOutputTokensEstimated ? "~" : "";
-      parts.push(chalk2.dim("in:") + chalk2.green(` ${prefix}${outTokens}`));
+      parts.push(chalk.dim("in:") + chalk.green(` ${prefix}${outTokens}`));
+    }
+    if (this.totalCost > 0) {
+      parts.push(chalk.dim("cost:") + chalk.cyan(` $${this.formatCost(this.totalCost)}`));
     }
-    parts.push(chalk2.dim(`${elapsed}s`));
-    this.target.write(`\r${chalk2.cyan(spinner)} ${parts.join(chalk2.dim(" | "))}`);
+    parts.push(chalk.dim(`${elapsed}s`));
+    this.target.write(`\r${chalk.cyan(spinner)} ${parts.join(chalk.dim(" | "))}`);
   }
   renderCumulativeMode(spinner) {
     const elapsed = ((Date.now() - this.totalStartTime) / 1e3).toFixed(1);
     const parts = [];
     if (this.model) {
-      parts.push(chalk2.cyan(this.model));
+      parts.push(chalk.cyan(this.model));
     }
     if (this.totalTokens > 0) {
-      parts.push(chalk2.dim("total:") + chalk2.magenta(` ${this.totalTokens}`));
+      parts.push(chalk.dim("total:") + chalk.magenta(` ${this.totalTokens}`));
     }
     if (this.iterations > 0) {
-      parts.push(chalk2.dim("iter:") + chalk2.blue(` ${this.iterations}`));
+      parts.push(chalk.dim("iter:") + chalk.blue(` ${this.iterations}`));
     }
-    parts.push(chalk2.dim(`${elapsed}s`));
-    this.target.write(`\r${chalk2.cyan(spinner)} ${parts.join(chalk2.dim(" | "))}`);
+    if (this.totalCost > 0) {
+      parts.push(chalk.dim("cost:") + chalk.cyan(` $${this.formatCost(this.totalCost)}`));
+    }
+    parts.push(chalk.dim(`${elapsed}s`));
+    this.target.write(`\r${chalk.cyan(spinner)} ${parts.join(chalk.dim(" | "))}`);
   }
   /**
    * Pauses the progress indicator and clears the line.
@@ -532,6 +555,12 @@ var StreamProgress = class {
   complete() {
     this.pause();
   }
+  /**
+   * Returns the total accumulated cost across all calls.
+   */
+  getTotalCost() {
+    return this.totalCost;
+  }
   /**
    * Returns a formatted prompt string with stats (like bash PS1).
    * Shows current call stats during streaming, cumulative stats otherwise.
@@ -546,25 +575,28 @@ var StreamProgress = class {
       if (this.callInputTokens > 0) {
         const prefix = this.callInputTokensEstimated ? "~" : "";
         parts.push(
-          chalk2.dim("out:") + chalk2.yellow(` ${prefix}${this.formatTokens(this.callInputTokens)}`)
+          chalk.dim("out:") + chalk.yellow(` ${prefix}${this.formatTokens(this.callInputTokens)}`)
         );
       }
       if (outTokens > 0) {
         const prefix = outEstimated ? "~" : "";
-        parts.push(chalk2.dim("in:") + chalk2.green(` ${prefix}${this.formatTokens(outTokens)}`));
+        parts.push(chalk.dim("in:") + chalk.green(` ${prefix}${this.formatTokens(outTokens)}`));
       }
-      parts.push(chalk2.dim(`${elapsed}s`));
+      parts.push(chalk.dim(`${elapsed}s`));
     } else {
       const elapsed = Math.round((Date.now() - this.totalStartTime) / 1e3);
       if (this.totalTokens > 0) {
-        parts.push(chalk2.magenta(this.formatTokens(this.totalTokens)));
+        parts.push(chalk.magenta(this.formatTokens(this.totalTokens)));
       }
       if (this.iterations > 0) {
-        parts.push(chalk2.blue(`i${this.iterations}`));
+        parts.push(chalk.blue(`i${this.iterations}`));
+      }
+      if (this.totalCost > 0) {
+        parts.push(chalk.cyan(`$${this.formatCost(this.totalCost)}`));
       }
-      parts.push(chalk2.dim(`${elapsed}s`));
+      parts.push(chalk.dim(`${elapsed}s`));
     }
-    return `${parts.join(chalk2.dim(" \u2502 "))} ${chalk2.green(">")} `;
+    return `${parts.join(chalk.dim(" \u2502 "))} ${chalk.green(">")} `;
   }
   /**
    * Formats token count compactly (3625 -> "3.6k").
@@ -572,6 +604,21 @@ var StreamProgress = class {
   formatTokens(tokens) {
     return tokens >= 1e3 ? `${(tokens / 1e3).toFixed(1)}k` : `${tokens}`;
   }
+  /**
+   * Formats cost compactly (0.0001234 -> "0.00012", 0.1234 -> "0.12", 1.234 -> "1.23").
+   */
+  formatCost(cost) {
+    if (cost < 1e-3) {
+      return cost.toFixed(5);
+    }
+    if (cost < 0.01) {
+      return cost.toFixed(4);
+    }
+    if (cost < 1) {
+      return cost.toFixed(3);
+    }
+    return cost.toFixed(2);
+  }
 };
 async function readStream(stream) {
   const chunks = [];
@@ -603,29 +650,42 @@ async function resolvePrompt(promptArg, env) {
 function renderSummary(metadata) {
   const parts = [];
   if (metadata.iterations !== void 0) {
-    parts.push(chalk2.dim(`iterations: ${metadata.iterations}`));
+    parts.push(chalk.dim(`iterations: ${metadata.iterations}`));
   }
   if (metadata.finishReason) {
-    parts.push(chalk2.dim(`finish: ${metadata.finishReason}`));
+    parts.push(chalk.dim(`finish: ${metadata.finishReason}`));
   }
   if (metadata.usage) {
     const { inputTokens, outputTokens, totalTokens } = metadata.usage;
     parts.push(
-      chalk2.dim(`tokens: `) + chalk2.cyan(`${totalTokens}`) + chalk2.dim(` (in: ${inputTokens}, out: ${outputTokens})`)
+      chalk.dim(`tokens: `) + chalk.cyan(`${totalTokens}`) + chalk.dim(` (in: ${inputTokens}, out: ${outputTokens})`)
     );
   }
+  if (metadata.cost !== void 0 && metadata.cost > 0) {
+    let formattedCost;
+    if (metadata.cost < 1e-3) {
+      formattedCost = metadata.cost.toFixed(5);
+    } else if (metadata.cost < 0.01) {
+      formattedCost = metadata.cost.toFixed(4);
+    } else if (metadata.cost < 1) {
+      formattedCost = metadata.cost.toFixed(3);
+    } else {
+      formattedCost = metadata.cost.toFixed(2);
+    }
+    parts.push(chalk.dim(`cost: `) + chalk.cyan(`$${formattedCost}`));
+  }
   if (parts.length === 0) {
     return null;
   }
-  return `${chalk2.dim("\u2500".repeat(40))}
-${parts.join(chalk2.dim(" \u2502 "))}`;
+  return `${chalk.dim("\u2500".repeat(40))}
+${parts.join(chalk.dim(" \u2502 "))}`;
 }
 async function executeAction(action, env) {
   try {
     await action();
   } catch (error) {
     const message = error instanceof Error ? error.message : String(error);
-    env.stderr.write(`${chalk2.red.bold("Error:")} ${message}
+    env.stderr.write(`${chalk.red.bold("Error:")} ${message}
 `);
     env.setExitCode(1);
   }
@@ -669,17 +729,18 @@ ${statsPrompt}` : statsPrompt;
   };
 }
 function formatGadgetSummary(result) {
-  const gadgetLabel = chalk3.magenta.bold(result.gadgetName);
-  const timeLabel = chalk3.dim(`${Math.round(result.executionTimeMs)}ms`);
+  const gadgetLabel = chalk2.magenta.bold(result.gadgetName);
+  const timeLabel = chalk2.dim(`${Math.round(result.executionTimeMs)}ms`);
   if (result.error) {
-    return `${chalk3.red("\u2717")} ${gadgetLabel} ${chalk3.red("error:")} ${result.error} ${timeLabel}`;
+    return `${chalk2.red("\u2717")} ${gadgetLabel} ${chalk2.red("error:")} ${result.error} ${timeLabel}`;
   }
   if (result.breaksLoop) {
-    return `${chalk3.yellow("\u23F9")} ${gadgetLabel} ${chalk3.yellow("finished:")} ${result.result} ${timeLabel}`;
+    return `${chalk2.yellow("\u23F9")} ${gadgetLabel} ${chalk2.yellow("finished:")} ${result.result} ${timeLabel}`;
   }
   const maxLen = 80;
-  const resultText = result.result ? result.result.length > maxLen ? `${result.result.slice(0, maxLen)}...` : result.result : "";
-  return `${chalk3.green("\u2713")} ${gadgetLabel} ${chalk3.dim("\u2192")} ${resultText} ${timeLabel}`;
+  const shouldTruncate = result.gadgetName !== "TellUser";
+  const resultText = result.result ? shouldTruncate && result.result.length > maxLen ? `${result.result.slice(0, maxLen)}...` : result.result : "";
+  return `${chalk2.green("\u2713")} ${gadgetLabel} ${chalk2.dim("\u2192")} ${resultText} ${timeLabel}`;
 }
 async function handleAgentCommand(promptArg, options, env) {
   const prompt = await resolvePrompt(promptArg, env);
@@ -699,7 +760,7 @@ async function handleAgentCommand(promptArg, options, env) {
   }
   const printer = new StreamPrinter(env.stdout);
   const stderrTTY = env.stderr.isTTY === true;
-  const progress = new StreamProgress(env.stderr, stderrTTY);
+  const progress = new StreamProgress(env.stderr, stderrTTY, client.modelRegistry);
   let finishReason;
   let usage;
   let iterations = 0;
@@ -756,16 +817,25 @@ async function handleAgentCommand(promptArg, options, env) {
       printer.write(event.content);
     } else if (event.type === "gadget_result") {
       progress.pause();
-      env.stderr.write(`${formatGadgetSummary(event.result)}
+      if (stderrTTY) {
+        env.stderr.write(`${formatGadgetSummary(event.result)}
 `);
+      }
     }
   }
   progress.complete();
   printer.ensureNewline();
-  const summary = renderSummary({ finishReason, usage, iterations });
-  if (summary) {
-    env.stderr.write(`${summary}
+  if (stderrTTY) {
+    const summary = renderSummary({
+      finishReason,
+      usage,
+      iterations,
+      cost: progress.getTotalCost()
+    });
+    if (summary) {
+      env.stderr.write(`${summary}
 `);
+    }
   }
 }
 function registerAgentCommand(program, env) {
@@ -794,26 +864,28 @@ function registerAgentCommand(program, env) {
 
 // src/cli/complete-command.ts
 init_messages();
+init_model_shortcuts();
 init_constants();
 async function handleCompleteCommand(promptArg, options, env) {
   const prompt = await resolvePrompt(promptArg, env);
   const client = env.createClient();
+  const model = resolveModel(options.model);
   const builder = new LLMMessageBuilder();
   if (options.system) {
     builder.addSystem(options.system);
   }
   builder.addUser(prompt);
   const stream = client.stream({
-    model: options.model,
+    model,
     messages: builder.build(),
     temperature: options.temperature,
     maxTokens: options.maxTokens
   });
   const printer = new StreamPrinter(env.stdout);
   const stderrTTY = env.stderr.isTTY === true;
-  const progress = new StreamProgress(env.stderr, stderrTTY);
+  const progress = new StreamProgress(env.stderr, stderrTTY, client.modelRegistry);
   const estimatedInputTokens = Math.round(prompt.length / FALLBACK_CHARS_PER_TOKEN);
-  progress.startCall(options.model, estimatedInputTokens);
+  progress.startCall(model, estimatedInputTokens);
   let finishReason;
   let usage;
   let totalChars = 0;
@@ -837,12 +909,15 @@ async function handleCompleteCommand(promptArg, options, env) {
       finishReason = chunk.finishReason;
     }
   }
+  progress.endCall(usage);
   progress.complete();
   printer.ensureNewline();
-  const summary = renderSummary({ finishReason, usage });
-  if (summary) {
-    env.stderr.write(`${summary}
+  if (stderrTTY) {
+    const summary = renderSummary({ finishReason, usage, cost: progress.getTotalCost() });
+    if (summary) {
+      env.stderr.write(`${summary}
 `);
+    }
   }
 }
 function registerCompleteCommand(program, env) {
@@ -866,7 +941,7 @@ function registerCompleteCommand(program, env) {
 init_client();
 init_logger();
 import readline from "node:readline";
-import chalk4 from "chalk";
+import chalk3 from "chalk";
 var LOG_LEVEL_MAP = {
   silly: 0,
   trace: 1,
@@ -910,14 +985,14 @@ function createPromptFunction(stdin, stdout) {
         output: stdout
       });
       stdout.write("\n");
-      stdout.write(`${chalk4.cyan("\u2500".repeat(60))}
+      stdout.write(`${chalk3.cyan("\u2500".repeat(60))}
 `);
-      stdout.write(chalk4.cyan.bold("\u{1F916} Agent asks:\n"));
+      stdout.write(chalk3.cyan.bold("\u{1F916} Agent asks:\n"));
       stdout.write(`${question}
 `);
-      stdout.write(`${chalk4.cyan("\u2500".repeat(60))}
+      stdout.write(`${chalk3.cyan("\u2500".repeat(60))}
 `);
-      rl.question(chalk4.green.bold("You: "), (answer) => {
+      rl.question(chalk3.green.bold("You: "), (answer) => {
         rl.close();
         resolve(answer);
       });