miii-agent 0.1.22 → 0.1.24

package/dist/cli.js

@@ -163,30 +163,6 @@ async function modelContext(entry, model) {
     throw err;
   }
 }
-async function paramCountB(entry, model) {
-  try {
-    const info = await makeClient(entry).show({ model });
-    const details = info.details;
-    if (details?.parameter_size) {
-      const m = details.parameter_size.match(/([\d.]+)\s*([BM])/i);
-      if (m) {
-        const n = parseFloat(m[1]);
-        if (!isNaN(n)) return m[2].toUpperCase() === "M" ? n / 1e3 : n;
-      }
-    }
-    const modelInfo = info.model_info;
-    if (modelInfo) {
-      const key = Object.keys(modelInfo).find((k) => k.endsWith("parameter_count"));
-      if (key) {
-        const val = Number(modelInfo[key]);
-        if (!isNaN(val) && val > 0) return val / 1e9;
-      }
-    }
-    return null;
-  } catch {
-    return null;
-  }
-}
 async function* chat(entry, model, messages, tools, opts) {
   if (opts?.signal?.aborted) return;
   const signal = opts?.signal;
@@ -214,8 +190,7 @@ async function* chat(entry, model, messages, tools, opts) {
       keep_alive: opts?.keep_alive ?? "10m",
       options
     };
-    if (opts?.format) req.format = opts.format;
-    else if (tools) req.tools = tools;
+    if (tools) req.tools = tools;
     try {
       stream = await client.chat(
         req
@@ -508,9 +483,6 @@ var init_openai = __esm({
 function active() {
   return resolveProvider();
 }
-function providerName() {
-  return active().name;
-}
 function isAvailable3() {
   const { entry } = active();
   return entry.type === "ollama" ? isAvailable(entry) : isAvailable2(entry);
@@ -527,16 +499,6 @@ async function modelContext3(model) {
   const { entry } = active();
   return entry.type === "ollama" ? modelContext(entry, model) : modelContext2(entry, model);
 }
-async function modelParamCountB(model) {
-  const { entry } = active();
-  if (entry.type !== "ollama") return null;
-  const key = `${entry.baseUrl}:${model}`;
-  const cached = paramCountCache.get(key);
-  if (cached !== void 0) return cached;
-  const params = await paramCountB(entry, model);
-  paramCountCache.set(key, params);
-  return params;
-}
 async function* chat3(model, messages, tools, opts) {
   const { entry } = active();
   if (entry.type === "ollama") {
@@ -545,70 +507,12 @@ async function* chat3(model, messages, tools, opts) {
     yield* chat2(entry, model, messages, tools, opts);
   }
 }
-var paramCountCache;
 var init_client = __esm({
   "src/llm/client.ts"() {
     "use strict";
     init_config();
     init_ollama();
     init_openai();
-    paramCountCache = /* @__PURE__ */ new Map();
-  }
-});
-
-// src/llm/grammar.ts
-function argProperties(props) {
-  const out = {};
-  for (const [key, spec] of Object.entries(props)) {
-    const node = { type: spec.type };
-    if (spec.enum && spec.enum.length) node.enum = spec.enum;
-    out[key] = node;
-  }
-  return out;
-}
-function toolBranch(tool) {
-  const args2 = {
-    type: "object",
-    additionalProperties: false,
-    properties: argProperties(tool.input_schema.properties)
-  };
-  if (tool.input_schema.required && tool.input_schema.required.length) {
-    args2.required = tool.input_schema.required;
-  }
-  return {
-    type: "object",
-    additionalProperties: false,
-    required: ["name", "arguments"],
-    properties: {
-      name: { const: tool.name },
-      arguments: args2
-    }
-  };
-}
-function respondBranch() {
-  return {
-    type: "object",
-    additionalProperties: false,
-    required: ["name", "arguments"],
-    properties: {
-      name: { const: RESPOND_ACTION },
-      arguments: {
-        type: "object",
-        additionalProperties: false,
-        required: ["message"],
-        properties: { message: { type: "string" } }
-      }
-    }
-  };
-}
-function buildToolGrammar(tools) {
-  return { oneOf: [...tools.map(toolBranch), respondBranch()] };
-}
-var RESPOND_ACTION;
-var init_grammar = __esm({
-  "src/llm/grammar.ts"() {
-    "use strict";
-    RESPOND_ACTION = "respond";
   }
 });
 
@@ -1304,15 +1208,8 @@ var init_context = __esm({
 });
 
 // src/prompt/system.ts
-function buildSystemPrompt(tools, cwd, project, grammarMode = false) {
+function buildSystemPrompt(tools, cwd, project) {
   const toolLines = tools.map((t) => `- ${t.name}: ${t.description}`).join("\n");
-  const actionProtocol = grammarMode ? `
-# Action protocol (strict)
-Every reply is exactly ONE JSON action object, nothing else \u2014 no prose outside it, no markdown, no fences. Decoding is grammar-constrained, so malformed output is impossible; your only job is to choose the right action.
-  To use a tool: {"name": "<tool_name>", "arguments": { ...that tool's args }}
-  To give your final answer to the user: {"name": "respond", "arguments": {"message": "<your full answer here>"}}
-Call tools until the GOAL is met, then emit a single "respond" action with the complete answer. The "respond" action is the ONLY way to end the turn and talk to the user \u2014 never put your final answer in a tool call.
-` : "";
   const projectSection = project && project.content.trim() ? `
 # ${CONTEXT_FILENAME} \u2014 project instructions (authoritative, read first)
 The user maintains ${CONTEXT_FILENAME} at ${project.source} to steer how you work in this project: conventions, commands, architecture, do's and don'ts. Treat it as direct instruction from the user, higher priority than your defaults. When it conflicts with a default rule below, ${CONTEXT_FILENAME} wins (except permissions and safety, which you never override).${project.truncated ? `
@@ -1397,7 +1294,7 @@ Ask in a numbered list. One round of questions per turn. Then wait.
 # Tools
 You have access to the following tools. Call them via the function-calling interface.
 ${toolLines}
-${actionProtocol}
+
 # Loop semantics
 - When you need to act on the filesystem or run a command, emit a tool call.
 - After each tool result, decide: more tool calls, or a final plain-text answer.
@@ -1627,89 +1524,6 @@ function extractFirstJsonObject(s) {
   }
   return null;
 }
-function parseGrammarAction(content, knownToolNames) {
-  if (!content) return null;
-  let raw = content.trim();
-  if (!raw.startsWith("{")) {
-    const found = extractFirstJsonObject(raw);
-    if (!found) return null;
-    raw = found.json;
-  }
-  let obj;
-  try {
-    obj = JSON.parse(raw);
-  } catch {
-    const found = extractFirstJsonObject(raw);
-    if (!found) return null;
-    try {
-      obj = JSON.parse(found.json);
-    } catch {
-      return null;
-    }
-  }
-  const name = typeof obj.name === "string" ? obj.name : void 0;
-  if (!name) return null;
-  let args2;
-  const wrapped = obj.arguments ?? obj.parameters ?? obj.input ?? obj.args;
-  if (typeof wrapped === "string") {
-    try {
-      const parsed = JSON.parse(wrapped);
-      args2 = parsed && typeof parsed === "object" && !Array.isArray(parsed) ? parsed : {};
-    } catch {
-      args2 = {};
-    }
-  } else if (wrapped && typeof wrapped === "object" && !Array.isArray(wrapped)) {
-    args2 = wrapped;
-  } else {
-    const { name: _n, ...rest } = obj;
-    args2 = rest;
-  }
-  if (name === "respond") {
-    const message = typeof args2.message === "string" ? args2.message : "";
-    return { kind: "respond", message };
-  }
-  if (!knownToolNames.includes(name)) return null;
-  return { kind: "tool", name, arguments: args2 };
-}
-function streamRespondMessage(text) {
-  if (!/"name"\s*:\s*"respond"/.test(text)) return null;
-  const m = text.match(/"message"\s*:\s*"/);
-  if (!m || m.index == null) return null;
-  const start = m.index + m[0].length;
-  const escapes = {
-    n: "\n",
-    t: "	",
-    r: "\r",
-    b: "\b",
-    f: "\f",
-    '"': '"',
-    "\\": "\\",
-    "/": "/"
-  };
-  let out = "";
-  let i = start;
-  while (i < text.length) {
-    const ch = text[i];
-    if (ch === '"') return { message: out, complete: true };
-    if (ch === "\\") {
-      const nx = text[i + 1];
-      if (nx === void 0) break;
-      if (nx === "u") {
-        const hex = text.slice(i + 2, i + 6);
-        if (hex.length < 4) break;
-        out += String.fromCharCode(parseInt(hex, 16));
-        i += 6;
-        continue;
-      }
-      out += escapes[nx] ?? nx;
-      i += 2;
-      continue;
-    }
-    out += ch;
-    i++;
-  }
-  return { message: out, complete: false };
-}
 function blocksFromOllama(text, tool_calls, knownToolNames = []) {
   const blocks = [];
   let finalText = text;
@@ -1791,13 +1605,7 @@ function markSeen(name, input, seen) {
 async function* runAgent(opts) {
   const { model, cwd, permissions, hooks, signal, num_ctx } = opts;
   const startTime = Date.now();
-  let useGrammar = false;
-  if (providerName() === "ollama") {
-    const params = await modelParamCountB(model);
-    useGrammar = params == null || params <= GRAMMAR_MAX_PARAMS_B;
-  }
-  const system = buildSystemPrompt(TOOLS, cwd, loadProjectContext(cwd), useGrammar);
-  const grammar = useGrammar ? buildToolGrammar(TOOLS) : void 0;
+  const system = buildSystemPrompt(TOOLS, cwd, loadProjectContext(cwd));
   const ollamaTools = toOllamaTools(TOOLS);
   const toolNames = TOOLS.map((t) => t.name);
   const effort = EFFORT_OPTIONS[loadConfig().effort ?? "medium"];
@@ -1813,8 +1621,6 @@ async function* runAgent(opts) {
   for (let turn = 0; turn < MAX_TURNS; turn++) {
     let text = "";
     let tool_calls;
-    let respondEmitted = 0;
-    let streamedRespond = false;
     let emittedText = false;
     let lastTail = "";
     let tailRepeats = 0;
@@ -1824,24 +1630,12 @@ async function* runAgent(opts) {
     const composedSignal = signal ? AbortSignal.any ? AbortSignal.any([signal, ac.signal]) : ac.signal : ac.signal;
     if (signal) signal.addEventListener("abort", () => ac.abort(), { once: true });
     try {
-      for await (const chunk of chat3(model, toOllamaMessages(history, system), useGrammar ? void 0 : ollamaTools, { signal: composedSignal, num_ctx, num_predict: effort.num_predict, temperature: effort.temperature, format: grammar })) {
+      for await (const chunk of chat3(model, toOllamaMessages(history, system), ollamaTools, { signal: composedSignal, num_ctx, num_predict: effort.num_predict, temperature: effort.temperature })) {
         if (signal?.aborted) break;
         if (chunk.content) {
           text += chunk.content;
-          if (!useGrammar) {
-            emittedText = true;
-            yield { type: "text-delta", text: chunk.content };
-          } else {
-            const r = streamRespondMessage(text);
-            if (r) {
-              streamedRespond = true;
-              if (r.message.length > respondEmitted) {
-                emittedText = true;
-                yield { type: "text-delta", text: r.message.slice(respondEmitted) };
-                respondEmitted = r.message.length;
-              }
-            }
-          }
+          emittedText = true;
+          yield { type: "text-delta", text: chunk.content };
           if (text.length >= REPEAT_TAIL) {
             const tail = text.slice(-REPEAT_TAIL);
             if (tail === lastTail) {
@@ -1890,19 +1684,7 @@ async function* runAgent(opts) {
       };
       return history;
     }
-    let blocks;
-    if (useGrammar) {
-      const action = parseGrammarAction(text, toolNames);
-      if (action?.kind === "tool") {
-        blocks = [{ type: "tool_use", id: mintToolUseId(), name: action.name, input: action.arguments }];
-      } else {
-        const message = action?.kind === "respond" ? action.message : text.trim();
-        if (message && !streamedRespond) yield { type: "text-delta", text: message };
-        blocks = message ? [{ type: "text", text: message }] : [];
-      }
-    } else {
-      blocks = blocksFromOllama(text, tool_calls, toolNames);
-    }
+    const blocks = blocksFromOllama(text, tool_calls, toolNames);
     const tool_uses = blocks.filter((b) => b.type === "tool_use");
     history.push({ role: "assistant", content: blocks });
     if (truncated && tool_uses.length > 0) {
@@ -2026,12 +1808,11 @@ async function* runAgent(opts) {
   yield { type: "done", prompt_tokens: promptTokens, eval_tokens: evalTokens };
   return history;
 }
-var MAX_TURNS, REPEAT_TAIL, REPEAT_KILL, GRAMMAR_MAX_PARAMS_B, BIG_WRITE_TOOLS;
+var MAX_TURNS, REPEAT_TAIL, REPEAT_KILL, BIG_WRITE_TOOLS;
 var init_loop = __esm({
   "src/agent/loop.ts"() {
     "use strict";
     init_client();
-    init_grammar();
     init_paths();
     init_registry();
     init_validate();
@@ -2043,7 +1824,6 @@ var init_loop = __esm({
     MAX_TURNS = 25;
     REPEAT_TAIL = 120;
     REPEAT_KILL = 4;
-    GRAMMAR_MAX_PARAMS_B = 14;
     BIG_WRITE_TOOLS = /* @__PURE__ */ new Set(["write_file", "edit_file"]);
   }
 });
@@ -2283,32 +2063,33 @@ import { Box, Text } from "ink";
 import { jsx, jsxs } from "react/jsx-runtime";
 function WelcomeBlock({ model, activeCtx, effort, cwd, updateAvailable }) {
   const ctxLabel = activeCtx != null ? `${Math.round(activeCtx / 1024)}k ctx` : "\u2014 ctx";
-  return /* @__PURE__ */ jsxs(
-    Box,
-    {
-      flexDirection: "column",
-      borderStyle: "round",
-      borderColor: "gray",
-      paddingX: 2,
-      marginBottom: 1,
-      children: [
-        /* @__PURE__ */ jsxs(Box, { gap: 2, children: [
-          /* @__PURE__ */ jsx(Text, { bold: true, color: "blue", children: "MIII CLI" }),
-          /* @__PURE__ */ jsx(Text, { dimColor: true, children: "\xB7" }),
-          /* @__PURE__ */ jsx(Text, { children: model ?? "/models" }),
-          /* @__PURE__ */ jsx(Text, { dimColor: true, children: "\xB7" }),
-          /* @__PURE__ */ jsx(Text, { children: ctxLabel }),
-          /* @__PURE__ */ jsx(Text, { dimColor: true, children: "\xB7" }),
-          /* @__PURE__ */ jsxs(Text, { children: [
-            effort,
-            " effort"
-          ] })
-        ] }),
-        /* @__PURE__ */ jsx(Text, { dimColor: true, children: cwd }),
-        updateAvailable && /* @__PURE__ */ jsx(Text, { color: "yellow", children: `\u2191 update available: v${updateAvailable} \u2014 run: miii --update` })
-      ]
-    }
-  );
+  return /* @__PURE__ */ jsxs(Box, { flexDirection: "column", marginBottom: 1, children: [
+    /* @__PURE__ */ jsxs(
+      Box,
+      {
+        flexDirection: "column",
+        borderStyle: "round",
+        borderColor: "gray",
+        paddingX: 2,
+        children: [
+          /* @__PURE__ */ jsxs(Box, { gap: 2, children: [
+            /* @__PURE__ */ jsx(Text, { bold: true, color: "blue", children: "MIII CLI" }),
+            /* @__PURE__ */ jsx(Text, { dimColor: true, children: "\xB7" }),
+            /* @__PURE__ */ jsx(Text, { children: model ?? "/models" }),
+            /* @__PURE__ */ jsx(Text, { dimColor: true, children: "\xB7" }),
+            /* @__PURE__ */ jsx(Text, { children: ctxLabel }),
+            /* @__PURE__ */ jsx(Text, { dimColor: true, children: "\xB7" }),
+            /* @__PURE__ */ jsxs(Text, { children: [
+              effort,
+              " effort"
+            ] })
+          ] }),
+          /* @__PURE__ */ jsx(Text, { dimColor: true, children: cwd })
+        ]
+      }
+    ),
+    updateAvailable && /* @__PURE__ */ jsx(Text, { color: "yellow", children: `\u2191 update available: v${updateAvailable} \u2014 run: miii --update` })
+  ] });
 }
 
 // src/ui/InputBar.tsx

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "miii-agent",
-  "version": "0.1.22",
+  "version": "0.1.24",
   "description": "Cursor / Claude Code, but local. An offline AI pair-programmer in your terminal, powered by Ollama. Private by default, free forever.",
   "type": "module",
   "bin": {