zapdev 0.11.0 → 1.0.0

package/README.md

@@ -4,7 +4,7 @@
 
 **zapdev** is a lightweight TypeScript CLI that makes small, repetitive Git chores fast and precise.
 
-It stages changes, scans them for secrets, generates Conventional Commit messages with Ollama, and streamlines repository cleanup.
+It stages changes, scans them for secrets, generates Conventional Commit messages with an OpenAI-compatible LLM endpoint, and streamlines repository cleanup.
 
 ## Project Architecture
 
@@ -16,24 +16,27 @@ flowchart LR
   Commands --> Types["src/types"]
   Lib --> Prompts["src/prompts"]
   Lib --> Types
-  Lib --> Tools["Git, Gitleaks, Ollama"]
+  Lib --> Tools["Git, Gitleaks, LLM API"]
 ```
 
 - `src/index.ts`: bin launcher; enables the V8 compile cache, then loads `cli.js`.
 - `src/cli.ts`: CLI entry; registers subcommands and opens the interactive menu.
 - `src/commands/`: command UI and orchestration.
-- `src/lib/`: pure logic and isolated Git, Gitleaks, and Ollama side effects.
+- `src/lib/`: pure logic and isolated Git, Gitleaks, and LLM side effects.
 - `src/prompts/`: LLM prompts inlined into the bundle at build time.
 - `src/types/`: shared type declarations.
 
 ## Environment Variables
 
-| Variable | Default | Required | Description |
-| --- | --- | --- | --- |
-| `OLLAMA_URL` | `http://localhost:11434` | No | Ollama base URL |
-| `OLLAMA_MODEL` | `deepseek-v4-flash:cloud` | No | Model used to generate commit messages |
-| `OLLAMA_BACKUP_MODEL` | - | No | Model used when generation with the primary model fails |
-| `OLLAMA_EFFORT` | `low` | No | Ollama thinking effort (`low`, `medium`, `high`, or `max`) |
+| Variable | Required | Description |
+| -------- | -------- | ----------- |
+| `ZD_URL` | For `commit` | Complete HTTP(S) Chat Completions endpoint, including its path |
+| `ZD_MODEL` | For `commit` | Model identifier supported by the endpoint |
+| `ZD_EFFORT` | For `commit` | Sent as `reasoning_effort`; use a value supported by your model, such as `low`, `medium`, or `high` |
+
+There are no defaults, automatic provider detection, or backup models. CLI flags override these variables. Legacy `OLLAMA_*` variables are no longer read. Other commands do not require LLM configuration.
+
+Requests use the OpenAI Chat Completions format over plain `fetch`, without a provider SDK. No authentication headers are sent; use an endpoint that does not require them. The endpoint and model must support `reasoning_effort`.
 
 ## Setup
 
@@ -41,7 +44,7 @@ flowchart LR
 
 - **Node.js >= 20 (required):** runs the CLI.
 - **Git (required):** provides the repository operations.
-- **Ollama (required for `commit`):** generates Conventional Commit messages.
+- **OpenAI-compatible Chat Completions endpoint (required for `commit`):** generates Conventional Commit messages.
 - **Gitleaks (recommended):** scans staged changes before message generation when available on `PATH`.
 
 ### Install
@@ -52,6 +55,14 @@ Install zapdev globally for daily use:
 npm install -g zapdev
 ```
 
+Configure your endpoint and model before running `commit` (replace these example values):
+
+```bash
+export ZD_URL="http://localhost:1234/v1/chat/completions"
+export ZD_MODEL="your-model-id"
+export ZD_EFFORT="low"
+```
+
 Or run it once without installing:
 
 ```bash
@@ -85,27 +96,61 @@ Stages all changes, scans them with Gitleaks when installed, generates a Convent
 zapdev commit
 ```
 
-| Flag | Description |
-| --- | --- |
-| `--model <model>` | Override the Ollama model |
+| Flag                | Description                                                       |
+| ------------------- | ----------------------------------------------------------------- |
+| `--url <url>`       | Override the complete Chat Completions endpoint                   |
+| `--model <model>`   | Override the model                                               |
+| `--effort <effort>` | Override the reasoning effort                                    |
 | `-t, --type <type>` | Force the Conventional Commit type (`feat`, `fix`, `chore`, etc.) |
-| `-p, --push` | Push after committing without asking |
-| `-s, --staged` | Commit only changes that are already staged |
-| `-r, --rebase` | Rebase on upstream if the push is rejected |
-| `-m, --merge` | Merge upstream if the push is rejected |
-| `-y, --yes` | Skip prompts and commit directly |
+| `-p, --push`        | Push after committing without asking                              |
+| `-s, --staged`      | Commit only changes that are already staged                       |
+| `-r, --rebase`      | Rebase on upstream if the push is rejected                        |
+| `-m, --merge`       | Merge upstream if the push is rejected                            |
+| `-y, --yes`         | Skip prompts and commit directly                                  |
 
 ```bash
 zapdev commit -t feat      # force the type
 zapdev commit --staged     # leave unstaged changes untouched
 ```
 
-Before contacting Ollama, zapdev runs `gitleaks git --staged` when Gitleaks is installed. A failed scan stops the commit; when Gitleaks is absent, the scan is skipped.
+Before contacting the LLM endpoint, zapdev runs `gitleaks git --staged` when Gitleaks is installed. A failed scan stops the commit; when Gitleaks is absent, the scan is skipped. The staged diff is sent to the configured endpoint, which may be remote.
 
 Pushing is optimistic, with no preliminary fetch. If the branch is behind upstream, `--rebase` runs `git pull --rebase`, while `--merge` runs `git pull --no-rebase --no-edit`; zapdev then retries once. Without either flag, interactive runs ask whether to rebase, merge, or quit. Runs using `--yes` or without a TTY must provide one of the flags.
 
 Without a TTY, zapdev commits automatically and only pushes when `--push` is set.
 
+### Zed IDE
+
+For a faster review and commit workflow, review and stage changes from Zed's Git panel, then run `zapdev commit -syp` from a task. The command commits only staged changes, skips prompts, and pushes the commit.
+
+Add the following tasks to `.zed/tasks.json`:
+
+```json
+[
+  {
+    "label": "Safely commit staged changes.",
+    "command": "zapdev commit -s",
+    "reveal": "always",
+    "hide": "on_success",
+    "reveal_target": "center"
+  }
+]
+```
+
+Add this entry to Zed's `keymap.json` to run the commit task with `ctrl-cmd-enter`:
+
+```json
+{
+  "context": "Pane",
+  "bindings": {
+    "ctrl-cmd-enter": [
+      "task::Spawn",
+      { "task_name": "Safely commit staged changes." }
+    ]
+  }
+}
+```
+
 ### `zapdev reset`
 
 Operates on a Git repository or the direct child repositories of a directory. It fetches and prunes, switches branch, then permanently removes other local branches and linked worktrees.
@@ -117,12 +162,12 @@ zapdev reset -p              # switch to the principal branch without prompting
 zapdev reset -t dev          # switch to dev or fall back to the principal branch
 ```
 
-| Flag | Description |
-| --- | --- |
-| `-p, --principal` | Switch every repo to its resolved principal branch (`origin/HEAD`) |
-| `-t, --target <branch>` | Switch to a target branch, falling back to the principal branch |
-| `--pull` | Pull the checked-out branch after reset without asking |
-| `-y, --yes` | Switch and delete without confirmation |
+| Flag                    | Description                                                        |
+| ----------------------- | ------------------------------------------------------------------ |
+| `-p, --principal`       | Switch every repo to its resolved principal branch (`origin/HEAD`) |
+| `-t, --target <branch>` | Switch to a target branch, falling back to the principal branch    |
+| `--pull`                | Pull the checked-out branch after reset without asking             |
+| `-y, --yes`             | Switch and delete without confirmation                             |
 
 Deletion is permanent. Branches are removed with `git branch -D`; worktrees are removed with `git worktree remove --force`. Without a TTY, pass `--yes` or the destructive step is refused. `node_modules` is never scanned.
 

package/dist/cli.js

@@ -1718,8 +1718,8 @@ function _getBuiltinFlags(long, short, userNames, userAliases) {
 // package.json
 var package_default = {
   name: "zapdev",
-  version: "0.11.0",
-  description: "Fast, performant and lightweight git chores and dev tools using your terminal and your own local LLM.",
+  version: "1.0.0",
+  description: "Fast, performant and lightweight git chores and dev tools using your terminal and an OpenAI-compatible LLM endpoint.",
   type: "module",
   bin: {
     zapdev: "./dist/index.js"
@@ -1735,7 +1735,6 @@ var package_default = {
     "cli",
     "commit",
     "conventional-commits",
-    "ollama",
     "llm",
     "ai"
   ],
@@ -1812,23 +1811,25 @@ function sanitizeCommitMessage(raw) {
 }
 
 // src/lib/config.ts
-var DEFAULT_OLLAMA_URL = "http://localhost:11434";
-var DEFAULT_MODEL = "deepseek-v4-flash:cloud";
-var DEFAULT_EFFORT = "low";
 function resolveConfig(env = process.env, overrides = {}) {
-  return {
-    ollamaUrl: normalizeBaseUrl(overrides.ollamaUrl ?? env.OLLAMA_URL ?? DEFAULT_OLLAMA_URL),
-    model: overrides.model ?? env.OLLAMA_MODEL ?? DEFAULT_MODEL,
-    backupModel: overrides.backupModel ?? optionalValue(env.OLLAMA_BACKUP_MODEL),
-    effort: overrides.effort ?? optionalValue(env.OLLAMA_EFFORT) ?? DEFAULT_EFFORT
+  const config = {
+    url: (overrides.url ?? env.ZD_URL)?.trim() ?? "",
+    model: (overrides.model ?? env.ZD_MODEL)?.trim() ?? "",
+    effort: (overrides.effort ?? env.ZD_EFFORT)?.trim() ?? ""
   };
-}
-function optionalValue(value) {
-  return value?.trim() || void 0;
-}
-function normalizeBaseUrl(raw) {
-  const trimmed = raw.trim().replace(/\/+$/, "");
-  return /^https?:\/\//.test(trimmed) ? trimmed : `http://${trimmed}`;
+  for (const [key, value] of Object.entries(config)) {
+    if (!value) throw new Error(`Set ZD_${key.toUpperCase()} or --${key} before generating a commit message.`);
+  }
+  let url;
+  try {
+    url = new URL(config.url);
+  } catch {
+    throw new Error("URL must be a complete HTTP(S) Chat Completions endpoint.");
+  }
+  if (url.protocol !== "http:" && url.protocol !== "https:") {
+    throw new Error("URL must be a complete HTTP(S) Chat Completions endpoint.");
+  }
+  return config;
 }
 
 // src/lib/git.ts
@@ -2330,65 +2331,49 @@ var commit_message_default = "You are zapdev, a CLI tool that creates commits qu
 // src/prompts/index.ts
 var COMMIT_SYSTEM_PROMPT = commit_message_default.trim();
 
-// src/lib/ollama.ts
+// src/lib/llm.ts
 var REQUEST_TIMEOUT_MS = 25e3;
-var KEEP_ALIVE = "30m";
 async function generateCommitMessage(diff, config, type) {
   const systemPrompt = type ? applyCommitType(COMMIT_SYSTEM_PROMPT, type) : COMMIT_SYSTEM_PROMPT;
   const messages = [
     { role: "system", content: systemPrompt },
     { role: "user", content: truncateDiff(diff) }
   ];
-  let content;
-  try {
-    content = await chat(config, config.model, messages);
-  } catch (error) {
-    if (!config.backupModel || config.backupModel === config.model) throw error;
-    try {
-      content = await chat(config, config.backupModel, messages);
-    } catch (backupError) {
-      throw new AggregateError(
-        [error, backupError],
-        `Primary model failed: ${errorMessage(error)} Backup model failed: ${errorMessage(backupError)}`,
-        { cause: backupError }
-      );
-    }
-  }
-  return sanitizeCommitMessage(content);
-}
-async function chat(config, model, messages) {
-  const url = `${config.ollamaUrl}/api/chat`;
   let response;
+  let body;
   try {
-    response = await fetch(url, {
+    response = await fetch(config.url, {
       method: "POST",
       headers: { "content-type": "application/json" },
       body: JSON.stringify({
-        model,
+        model: config.model,
         stream: false,
-        keep_alive: KEEP_ALIVE,
-        think: config.effort,
-        options: { temperature: 0.2 },
+        reasoning_effort: config.effort,
         messages
       }),
       signal: AbortSignal.timeout(REQUEST_TIMEOUT_MS)
     });
+    const text2 = await response.text();
+    try {
+      body = JSON.parse(text2);
+    } catch {
+      body = null;
+    }
   } catch (error) {
-    throw new Error(describeRequestError(error, url), { cause: error });
+    throw new Error(describeRequestError(error), { cause: error });
   }
-  const body = await response.json().catch(() => null);
   if (!response.ok) {
-    throw new Error(`Ollama error: ${serverError(body) ?? `HTTP ${response.status}`}`);
+    throw new Error(`LLM error: ${serverError(body) ?? `HTTP ${response.status}`}`);
   }
   const content = messageContent(body);
-  if (content === null) throw new Error("Ollama returned an unexpected response shape.");
-  return content;
+  if (content === null) throw new Error("LLM returned an unexpected response shape.");
+  return sanitizeCommitMessage(content);
 }
-function describeRequestError(error, url) {
+function describeRequestError(error) {
   if (errorName(error) === "TimeoutError") {
-    return `Ollama did not answer within ${REQUEST_TIMEOUT_MS / 1e3}s (${url}).`;
+    return `LLM did not answer within ${REQUEST_TIMEOUT_MS / 1e3}s.`;
   }
-  return `Could not reach Ollama at ${url}. Is it running?`;
+  return "Could not complete the LLM request. Check URL and endpoint availability.";
 }
 function errorName(error) {
   if (error && typeof error === "object" && "name" in error && typeof error.name === "string") {
@@ -2397,14 +2382,18 @@ function errorName(error) {
   return null;
 }
 function serverError(body) {
-  if (body && typeof body === "object" && "error" in body && typeof body.error === "string") {
-    return body.error;
+  if (!body || typeof body !== "object" || !("error" in body)) return null;
+  const error = body.error;
+  if (error && typeof error === "object" && "message" in error && typeof error.message === "string") {
+    return error.message;
   }
   return null;
 }
 function messageContent(body) {
-  if (!body || typeof body !== "object" || !("message" in body)) return null;
-  const message = body.message;
+  if (!body || typeof body !== "object" || !("choices" in body) || !Array.isArray(body.choices)) return null;
+  const choice = body.choices[0];
+  if (!choice || typeof choice !== "object" || !("message" in choice)) return null;
+  const message = choice.message;
   if (!message || typeof message !== "object" || !("content" in message)) return null;
   return typeof message.content === "string" ? message.content : null;
 }
@@ -2416,9 +2405,17 @@ var commitCommand = defineCommand({
     description: "Stage all changes and commit with an LLM-generated Conventional Commits message."
   },
   args: {
+    url: {
+      type: "string",
+      description: "Override $ZD_URL, the complete Chat Completions endpoint."
+    },
     model: {
       type: "string",
-      description: "Override the Ollama model (defaults to $OLLAMA_MODEL)."
+      description: "Override the model configured with $ZD_MODEL."
+    },
+    effort: {
+      type: "string",
+      description: "Override $ZD_EFFORT, sent as reasoning_effort."
     },
     type: {
       type: "string",
@@ -2453,10 +2450,6 @@ var commitCommand = defineCommand({
   },
   async run({ args }) {
     const interactive = Boolean(process.stdin.isTTY && process.stdout.isTTY);
-    const config = resolveConfig(
-      process.env,
-      args.model ? { model: args.model } : {}
-    );
     if (args.rebase && args.merge) {
       log.error("Choose either --rebase or --merge, not both.");
       process.exitCode = 1;
@@ -2471,6 +2464,18 @@ var commitCommand = defineCommand({
       process.exitCode = 1;
       return;
     }
+    let config;
+    try {
+      config = resolveConfig(process.env, {
+        url: args.url,
+        model: args.model,
+        effort: args.effort
+      });
+    } catch (error) {
+      log.error(errorMessage(error));
+      process.exitCode = 1;
+      return;
+    }
     if (interactive) intro("zapdev commit");
     if (!args.staged) await stageAll();
     const diff = await getStagedDiff();

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "zapdev",
-  "version": "0.11.0",
-  "description": "Fast, performant and lightweight git chores and dev tools using your terminal and your own local LLM.",
+  "version": "1.0.0",
+  "description": "Fast, performant and lightweight git chores and dev tools using your terminal and an OpenAI-compatible LLM endpoint.",
   "type": "module",
   "bin": {
     "zapdev": "./dist/index.js"
@@ -17,7 +17,6 @@
     "cli",
     "commit",
     "conventional-commits",
-    "ollama",
     "llm",
     "ai"
   ],