oh-pi 0.1.0 → 0.1.2

package/README.md

@@ -1,179 +1,193 @@
-# oh-pi!
+<div align="center">
 
-> One-click setup for [pi-coding-agent](https://github.com/badlogic/pi-mono). Like oh-my-zsh for pi.
+<img src="./logo.svg" width="180" alt="oh-pi logo"/>
+
+# 🐜 oh-pi
+
+**One command to supercharge [pi-coding-agent](https://github.com/badlogic/pi-mono).**
+
+Like oh-my-zsh for pi — but with an autonomous ant colony.
+
+[![npm](https://img.shields.io/npm/v/oh-pi)](https://www.npmjs.com/package/oh-pi)
+[![license](https://img.shields.io/npm/l/oh-pi)](./LICENSE)
+[![node](https://img.shields.io/node/v/oh-pi)](https://nodejs.org)
 
 ```bash
 npx oh-pi
 ```
 
-## What it does
+</div>
 
-oh-pi! is a modern interactive TUI that configures pi-coding-agent in minutes:
+---
 
-- **API Key Setup** — Multi-provider configuration with validation (Anthropic, OpenAI, Google, Groq, OpenRouter, xAI, Mistral)
-- **Preset Profiles** — Pre-made configs for different roles (Developer, Security, Data/AI, Colony Operator, Minimal)
-- **Custom Themes** — 6 beautiful themes (oh-pi Dark, Cyberpunk, Nord, Catppuccin, Tokyo Night, Gruvbox)
-- **Prompt Templates** — 10 ready-to-use templates (/review, /fix, /commit, /test, /security, etc.)
-- **Extensions** — Safety guards, git checkpoints, auto session naming, ant colony swarm
-- **Skills** — Debug helper, git workflow, quick project setup, ant colony orchestration
-- **Keybindings** — Default, Vim, or Emacs schemes
-- **AGENTS.md** — Role-specific project guidelines
-- **🐜 Ant Colony** — Autonomous multi-agent swarm with adaptive concurrency
+## Why
+
+pi-coding-agent is powerful out of the box. But configuring providers, themes, extensions, skills, and prompts by hand is tedious. oh-pi gives you a modern TUI that does it all in under a minute — and ships an **ant colony swarm** that turns pi into a multi-agent system.
 
 ## Quick Start
 
 ```bash
-# Run the configurator
-npx oh-pi
+npx oh-pi    # configure everything
+pi           # start coding
+```
+
+That's it. oh-pi detects your environment, walks you through setup, and writes `~/.pi/agent/` for you.
 
-# Then start coding
-pi
+Already have a config? oh-pi detects it and offers **backup before overwriting**.
+
+## What You Get
+
+```
+~/.pi/agent/
+├── auth.json            API keys (0600 permissions)
+├── settings.json        Model, theme, thinking level
+├── keybindings.json     Vim/Emacs shortcuts (optional)
+├── AGENTS.md            Role-specific AI guidelines
+├── extensions/          4 extensions
+│   ├── safe-guard       Dangerous command confirmation + path protection
+│   ├── git-guard        Auto stash checkpoints + dirty repo warning
+│   ├── auto-session     Session naming from first message
+│   └── ant-colony/      🐜 Autonomous multi-agent swarm
+├── prompts/             10 templates (/review /fix /commit /test ...)
+├── skills/              4 skills (debug, git, setup, colony)
+└── themes/              6 custom themes
 ```
 
 ## Setup Modes
 
-### 🚀 Quick Setup (3 steps)
-1. Pick your API provider(s)
-2. Enter API key(s)
-3. Done — sensible defaults applied
+| Mode | Steps | For |
+|------|-------|-----|
+| 🚀 **Quick** | 3 | Pick provider → enter key → done |
+| 📦 **Preset** | 2 | Choose a role profile → enter key |
+| 🎛️ **Custom** | 6 | Pick everything yourself |
 
-### 📦 Preset
-Choose a pre-made profile:
+### Presets
 
-| Preset | Theme | Thinking | Focus |
-|--------|-------|----------|-------|
-| 🟢 Starter | oh-pi Dark | medium | Basic safety + git |
+| | Theme | Thinking | Includes |
+|---|-------|----------|----------|
+| 🟢 Starter | oh-pi Dark | medium | Safety + git basics |
 | 🔵 Pro Developer | Catppuccin | high | Full toolchain |
 | 🟣 Security Researcher | Cyberpunk | high | Audit + pentesting |
-| 🟠 Data & AI Engineer | Tokyo Night | medium | MLOps + pipelines |
-| 🔴 Minimal | Pi Default | off | Core only |
+| 🟠 Data & AI | Tokyo Night | medium | MLOps + pipelines |
+| 🔴 Minimal | Default | off | Core only |
 | ⚫ Full Power | oh-pi Dark | high | Everything + ant colony |
 
-### 🎛️ Custom
-Pick every option yourself: providers, theme, keybindings, extensions, skills, AGENTS.md template.
+### Providers
 
-## 🐜 Ant Colony
+Anthropic · OpenAI · Google Gemini · Groq · OpenRouter · xAI · Mistral
+
+Auto-detects API keys from environment variables.
 
-Autonomous multi-agent swarm built as a pi extension. Modeled after real ant colony behavior.
+## 🐜 Ant Colony
 
-### How it works
+The headline feature. A multi-agent swarm modeled after real ant ecology.
 
 ```
-Goal → 🔍 Scouts explore → 📋 Task pool generated → ⚒️ Workers execute in parallel → 🛡️ Soldiers review → ✅ Done
+You: "Refactor auth from sessions to JWT"
+
+oh-pi:
+  🔍 Scout ants explore codebase (haiku — fast, cheap)
+  📋 Task pool generated from discoveries
+  ⚒️  Worker ants execute in parallel (sonnet — capable)
+  🛡️ Soldier ants review all changes (sonnet — thorough)
+  ✅ Done — summary report with metrics
 ```
 
-- **Scouts** (haiku) — Fast codebase recon, identify targets
-- **Workers** (sonnet) — Execute tasks, can spawn sub-tasks
-- **Soldiers** (sonnet) — Review quality, request fixes if needed
+### Why ants?
 
-### Key features
+Real ant colonies solve complex problems without central control. Each ant follows simple rules, communicates through **pheromone trails**, and the colony self-organizes. oh-pi maps this directly:
 
-- **Auto-trigger** — LLM automatically deploys colony for complex multi-file tasks
-- **Adaptive concurrency** — Starts at 1, explores throughput ceiling, stabilizes at optimal
-- **429 backoff** — Rate limits trigger exponential backoff (15s→30s→60s) + concurrency halving
-- **Pheromone communication** — Ants share discoveries via file-based pheromone trails (10min half-life)
-- **File locking** — One ant per file, blocked tasks auto-resume when locks release
+| Real Ants | oh-pi |
+|-----------|-------|
+| Scout finds food | Scout scans codebase, identifies targets |
+| Pheromone trail | `.ant-colony/pheromone.jsonl` — shared discoveries |
+| Worker carries food | Worker executes task on assigned files |
+| Soldier guards nest | Soldier reviews changes, requests fixes |
+| More food → more ants | More tasks → higher concurrency (auto-adapted) |
+| Pheromone evaporates | 10-minute half-life — stale info fades |
 
-### Usage
+### Auto-trigger
 
-```bash
-# LLM auto-triggers for complex tasks
-"Refactor the auth system from sessions to JWT"
+The LLM decides when to deploy the colony. You don't have to think about it:
 
-# Manual command
+- **≥3 files** need changes → colony
+- **Parallel workstreams** possible → colony
+- **Single file** change → direct execution (no colony overhead)
+
+Or trigger manually:
+
+```
 /colony migrate the entire project from CJS to ESM
+```
+
+### Adaptive Concurrency
+
+The colony automatically finds the optimal parallelism for your machine:
 
-# Shortcut
-Ctrl+Alt+A
 ```
+Cold start     →  1-2 ants (conservative)
+Exploration    →  +1 each wave, monitoring throughput
+Throughput ↓   →  lock optimal, stabilize
+CPU > 85%      →  reduce immediately
+429 rate limit →  halve concurrency + exponential backoff (15s→30s→60s)
+Tasks done     →  scale down to minimum
+```
+
+### File Safety
 
-## What Gets Installed
+One ant per file. Always. Conflicting tasks are automatically blocked and resume when locks release.
+
+## Themes
+
+| | |
+|---|---|
+| 🌙 **oh-pi Dark** | Cyan + purple, high contrast |
+| 🌙 **Cyberpunk** | Neon magenta + electric cyan |
+| 🌙 **Nord** | Arctic blue palette |
+| 🌙 **Catppuccin Mocha** | Pastel on dark |
+| 🌙 **Tokyo Night** | Blue + purple twilight |
+| 🌙 **Gruvbox Dark** | Warm retro tones |
+
+## Prompt Templates
 
 ```
-~/.pi/agent/
-├── auth.json          # API keys (0600 permissions)
-├── settings.json      # Model, theme, thinking level
-├── keybindings.json   # Vim/Emacs shortcuts (if selected)
-├── AGENTS.md          # Project guidelines for the AI
-├── extensions/        # Safety guards, git tools, ant colony
-├── prompts/           # /review, /fix, /commit, /test, etc.
-├── skills/            # debug-helper, git-workflow, ant-colony
-└── themes/            # Custom color themes
+/review    Code review: bugs, security, performance
+/fix       Fix errors with minimal changes
+/explain   Explain code, simple to detailed
+/refactor  Refactor preserving behavior
+/test      Generate tests
+/commit    Conventional Commit message
+/pr        Pull request description
+/security  OWASP security audit
+/optimize  Performance optimization
+/document  Generate documentation
 ```
 
-Existing config? oh-pi! detects it and offers backup before overwriting.
-
-## Included Resources
-
-### Themes
-
-| Theme | Style |
-|-------|-------|
-| oh-pi Dark | Cyan + Purple, high contrast |
-| Cyberpunk | Neon magenta + electric cyan |
-| Nord | Arctic blue palette |
-| Catppuccin Mocha | Pastel colors on dark |
-| Tokyo Night | Blue + purple twilight |
-| Gruvbox Dark | Warm retro tones |
-
-### Prompt Templates
-
-| Command | Description |
-|---------|-------------|
-| `/review` | Code review: bugs, security, performance |
-| `/fix` | Fix errors with minimal changes |
-| `/explain` | Explain code from simple to detailed |
-| `/refactor` | Refactor while preserving behavior |
-| `/test` | Generate tests for code |
-| `/commit` | Conventional Commit message |
-| `/pr` | Pull request description |
-| `/security` | OWASP security audit |
-| `/optimize` | Performance optimization |
-| `/document` | Generate documentation |
-
-### Extensions
-
-| Extension | Description |
-|-----------|-------------|
-| Safe Guard | Confirms dangerous commands (rm -rf, DROP, etc.) + protects .env, .git/ |
-| Git Guard | Auto stash checkpoints + dirty repo warning + completion notification |
-| Auto Session Name | Names sessions from first message |
-| 🐜 Ant Colony | Autonomous multi-agent swarm with adaptive concurrency |
-
-### Skills
-
-| Skill | Description |
-|-------|-------------|
-| `/skill:quick-setup` | Detect project type, generate .pi/ config |
-| `/skill:debug-helper` | Error analysis, log interpretation, profiling |
-| `/skill:git-workflow` | Branch strategy, PR workflow, conflict resolution |
-| `/skill:ant-colony` | Colony orchestration strategies and tuning |
-
-### AGENTS.md Templates
-
-| Template | Description |
-|----------|-------------|
+## AGENTS.md Templates
+
+| Template | Focus |
+|----------|-------|
 | General Developer | Universal coding guidelines |
-| Full-Stack Developer | Frontend + Backend + DB |
+| Full-Stack Developer | Frontend + backend + DB |
 | Security Researcher | Pentesting & audit |
 | Data & AI Engineer | MLOps & pipelines |
-| 🐜 Colony Operator | Ant swarm multi-agent orchestration |
+| 🐜 Colony Operator | Multi-agent orchestration |
 
 ## Also a Pi Package
 
-oh-pi! is also a pi package. Install just the resources without the configurator:
+Skip the configurator, just install the resources:
 
 ```bash
 pi install npm:oh-pi
 ```
 
-This adds all themes, prompts, skills, and extensions to your pi setup.
+Adds all themes, prompts, skills, and extensions to your existing pi setup.
 
 ## Requirements
 
-- Node.js >= 20
-- pi-coding-agent (installed automatically if missing)
+- Node.js ≥ 20
 - At least one LLM API key
+- pi-coding-agent (installed automatically if missing)
 
 ## License
 

package/dist/index.js

@@ -1,3 +1,4 @@
+import * as p from "@clack/prompts";
 import { welcome } from "./tui/welcome.js";
 import { selectMode } from "./tui/mode-select.js";
 import { setupProviders } from "./tui/provider-setup.js";
@@ -49,6 +50,34 @@ async function customFlow(env) {
     const keybindings = await selectKeybindings();
     const extensions = await selectExtensions();
     const agents = await selectAgents();
+    // Advanced: auto-compaction threshold
+    const wantAdvanced = await p.confirm({
+        message: "Configure advanced settings? (compaction threshold, etc.)",
+        initialValue: false,
+    });
+    if (p.isCancel(wantAdvanced)) {
+        p.cancel("Cancelled.");
+        process.exit(0);
+    }
+    let compactThreshold = 0.75;
+    if (wantAdvanced) {
+        const threshold = await p.text({
+            message: "Auto-compact when context reaches % of window (0-100):",
+            placeholder: "75",
+            initialValue: "75",
+            validate: (v) => {
+                const n = Number(v);
+                if (isNaN(n) || n < 10 || n > 100)
+                    return "Must be a number between 10 and 100";
+                return undefined;
+            },
+        });
+        if (p.isCancel(threshold)) {
+            p.cancel("Cancelled.");
+            process.exit(0);
+        }
+        compactThreshold = Number(threshold) / 100;
+    }
     return {
         providers,
         theme,
@@ -58,5 +87,6 @@ async function customFlow(env) {
         prompts: ["review", "fix", "explain", "commit", "test", "refactor", "optimize", "security", "document", "pr"],
         agents,
         thinking: "medium",
+        compactThreshold,
     };
 }

package/dist/tui/confirm-apply.js

@@ -12,6 +12,7 @@ export async function confirmApply(config, env) {
         `Theme:        ${chalk.cyan(config.theme)}`,
         `Keybindings:  ${chalk.cyan(config.keybindings)}`,
         `Thinking:     ${chalk.cyan(config.thinking)}`,
+        `Compaction:   ${chalk.cyan(`${Math.round((config.compactThreshold ?? 0.75) * 100)}% of context`)}`,
         `Extensions:   ${chalk.cyan(config.extensions.join(", ") || "none")}`,
         `Skills:       ${chalk.cyan(config.skills.join(", ") || "none")}`,
         `Prompts:      ${chalk.cyan(`${config.prompts.length} templates`)}`,

package/dist/tui/provider-setup.js

@@ -1,15 +1,31 @@
 import * as p from "@clack/prompts";
 import chalk from "chalk";
 import { PROVIDERS } from "../types.js";
+/** Fetch models from OpenAI-compatible /v1/models endpoint */
+async function fetchModels(baseUrl, apiKey) {
+    const url = `${baseUrl.replace(/\/+$/, "")}/v1/models`;
+    try {
+        const res = await fetch(url, {
+            headers: { Authorization: `Bearer ${apiKey}` },
+            signal: AbortSignal.timeout(8000),
+        });
+        if (!res.ok)
+            return [];
+        const json = await res.json();
+        return (json.data ?? []).map(m => m.id).sort();
+    }
+    catch {
+        return [];
+    }
+}
 export async function setupProviders() {
     const entries = Object.entries(PROVIDERS);
     const selected = await p.multiselect({
         message: "Select API providers",
-        options: entries.map(([key, info]) => ({
-            value: key,
-            label: info.label,
-            hint: info.env,
-        })),
+        options: [
+            ...entries.map(([key, info]) => ({ value: key, label: info.label, hint: info.env })),
+            { value: "_custom", label: "🔧 Custom endpoint", hint: "Ollama, vLLM, LiteLLM, any OpenAI-compatible" },
+        ],
         initialValues: ["anthropic"],
         required: true,
     });
@@ -19,45 +35,200 @@ export async function setupProviders() {
     }
     const configs = [];
     for (const name of selected) {
+        if (name === "_custom") {
+            const custom = await setupCustomProvider();
+            if (custom)
+                configs.push(custom);
+            continue;
+        }
         const info = PROVIDERS[name];
         const envVal = process.env[info.env];
         let apiKey;
         if (envVal) {
-            const useEnv = await p.confirm({
-                message: `Found ${chalk.cyan(info.env)} in environment. Use it?`,
-            });
+            const useEnv = await p.confirm({ message: `Found ${chalk.cyan(info.env)} in environment. Use it?` });
             if (p.isCancel(useEnv)) {
                 p.cancel("Cancelled.");
                 process.exit(0);
             }
-            apiKey = useEnv ? info.env : await promptKey(info);
+            apiKey = useEnv ? info.env : await promptKey(info.label);
         }
         else {
-            apiKey = await promptKey(info);
+            apiKey = await promptKey(info.label);
         }
-        let defaultModel;
-        if (info.models.length > 1) {
-            const model = await p.select({
-                message: `Default model for ${info.label}:`,
-                options: info.models.map(m => ({ value: m, label: m })),
+        // Ask for custom base URL (optional)
+        const wantCustomUrl = await p.confirm({
+            message: `Custom endpoint for ${info.label}? (proxy, Azure, etc.)`,
+            initialValue: false,
+        });
+        if (p.isCancel(wantCustomUrl)) {
+            p.cancel("Cancelled.");
+            process.exit(0);
+        }
+        let baseUrl;
+        if (wantCustomUrl) {
+            const url = await p.text({
+                message: `Base URL for ${info.label}:`,
+                placeholder: "https://your-proxy.example.com",
+                validate: (v) => (!v || !v.startsWith("http")) ? "Must be a valid URL" : undefined,
             });
-            if (p.isCancel(model)) {
+            if (p.isCancel(url)) {
                 p.cancel("Cancelled.");
                 process.exit(0);
             }
-            defaultModel = model;
-        }
-        else {
-            defaultModel = info.models[0];
+            baseUrl = url;
         }
-        configs.push({ name, apiKey, defaultModel });
+        // Model selection — try dynamic fetch, fall back to static list
+        const defaultModel = await selectModel(info.label, info.models, baseUrl, apiKey);
+        configs.push({ name, apiKey, defaultModel, ...(baseUrl ? { baseUrl } : {}) });
         p.log.success(`${info.label} configured`);
     }
     return configs;
 }
-async function promptKey(info) {
+async function setupCustomProvider() {
+    const name = await p.text({
+        message: "Provider name:",
+        placeholder: "ollama",
+        validate: (v) => (!v || v.trim().length === 0) ? "Name required" : undefined,
+    });
+    if (p.isCancel(name)) {
+        p.cancel("Cancelled.");
+        process.exit(0);
+    }
+    const baseUrl = await p.text({
+        message: "Base URL:",
+        placeholder: "http://localhost:11434",
+        validate: (v) => (!v || !v.startsWith("http")) ? "Must be a valid URL" : undefined,
+    });
+    if (p.isCancel(baseUrl)) {
+        p.cancel("Cancelled.");
+        process.exit(0);
+    }
+    const needsKey = await p.confirm({ message: "Requires API key?", initialValue: false });
+    if (p.isCancel(needsKey)) {
+        p.cancel("Cancelled.");
+        process.exit(0);
+    }
+    let apiKey = "none";
+    if (needsKey) {
+        apiKey = await promptKey(name);
+    }
+    // Dynamic model fetch
+    const s = p.spinner();
+    s.start(`Fetching models from ${baseUrl}`);
+    const models = await fetchModels(baseUrl, apiKey);
+    s.stop(models.length > 0 ? `Found ${models.length} models` : "No models found via API");
+    let defaultModel;
+    if (models.length > 0) {
+        const model = await p.select({
+            message: `Default model for ${name}:`,
+            options: models.slice(0, 30).map(m => ({ value: m, label: m })),
+        });
+        if (p.isCancel(model)) {
+            p.cancel("Cancelled.");
+            process.exit(0);
+        }
+        defaultModel = model;
+    }
+    else {
+        const model = await p.text({
+            message: `Model name for ${name}:`,
+            placeholder: "llama3.1:8b",
+            validate: (v) => (!v || v.trim().length === 0) ? "Model name required" : undefined,
+        });
+        if (p.isCancel(model)) {
+            p.cancel("Cancelled.");
+            process.exit(0);
+        }
+        defaultModel = model;
+    }
+    p.log.success(`${name} configured (${baseUrl})`);
+    // Model capabilities (optional)
+    const wantCaps = await p.confirm({
+        message: "Configure model capabilities? (context window, multimodal, reasoning)",
+        initialValue: false,
+    });
+    if (p.isCancel(wantCaps)) {
+        p.cancel("Cancelled.");
+        process.exit(0);
+    }
+    let contextWindow;
+    let maxTokens;
+    let reasoning;
+    let multimodal;
+    if (wantCaps) {
+        const ctxInput = await p.text({
+            message: "Context window size (tokens):",
+            placeholder: "128000",
+            initialValue: "128000",
+            validate: (v) => {
+                const n = Number(v);
+                if (isNaN(n) || n < 1024)
+                    return "Must be a number ≥ 1024";
+                return undefined;
+            },
+        });
+        if (p.isCancel(ctxInput)) {
+            p.cancel("Cancelled.");
+            process.exit(0);
+        }
+        contextWindow = Number(ctxInput);
+        const maxTokInput = await p.text({
+            message: "Max output tokens:",
+            placeholder: "8192",
+            initialValue: "8192",
+            validate: (v) => {
+                const n = Number(v);
+                if (isNaN(n) || n < 256)
+                    return "Must be a number ≥ 256";
+                return undefined;
+            },
+        });
+        if (p.isCancel(maxTokInput)) {
+            p.cancel("Cancelled.");
+            process.exit(0);
+        }
+        maxTokens = Number(maxTokInput);
+        const isMultimodal = await p.confirm({ message: "Supports image input (multimodal)?", initialValue: false });
+        if (p.isCancel(isMultimodal)) {
+            p.cancel("Cancelled.");
+            process.exit(0);
+        }
+        multimodal = isMultimodal;
+        const isReasoning = await p.confirm({ message: "Supports extended thinking (reasoning)?", initialValue: false });
+        if (p.isCancel(isReasoning)) {
+            p.cancel("Cancelled.");
+            process.exit(0);
+        }
+        reasoning = isReasoning;
+    }
+    return { name, apiKey, defaultModel, baseUrl, contextWindow, maxTokens, reasoning, multimodal };
+}
+async function selectModel(label, staticModels, baseUrl, apiKey) {
+    let models = staticModels;
+    // Try dynamic fetch if custom URL or known provider
+    if (baseUrl && apiKey) {
+        const s = p.spinner();
+        s.start(`Fetching models from ${label}`);
+        const fetched = await fetchModels(baseUrl, apiKey);
+        s.stop(fetched.length > 0 ? `Found ${fetched.length} models` : "Using default model list");
+        if (fetched.length > 0)
+            models = fetched;
+    }
+    if (models.length === 1)
+        return models[0];
+    const model = await p.select({
+        message: `Default model for ${label}:`,
+        options: models.slice(0, 30).map(m => ({ value: m, label: m })),
+    });
+    if (p.isCancel(model)) {
+        p.cancel("Cancelled.");
+        process.exit(0);
+    }
+    return model;
+}
+async function promptKey(label) {
     const key = await p.password({
-        message: `API key for ${info.label}:`,
+        message: `API key for ${label}:`,
         validate: (v) => (!v || v.trim().length === 0) ? "API key cannot be empty" : undefined,
     });
     if (p.isCancel(key)) {

package/dist/types.d.ts

@@ -2,6 +2,11 @@ export interface ProviderConfig {
     name: string;
     apiKey: string;
     defaultModel?: string;
+    baseUrl?: string;
+    contextWindow?: number;
+    maxTokens?: number;
+    reasoning?: boolean;
+    multimodal?: boolean;
 }
 export interface OhPConfig {
     providers: ProviderConfig[];
@@ -12,7 +17,16 @@ export interface OhPConfig {
     prompts: string[];
     agents: string;
     thinking: string;
+    compactThreshold?: number;
 }
+/** Official model capabilities for known providers */
+export interface ModelCapabilities {
+    contextWindow: number;
+    maxTokens: number;
+    reasoning: boolean;
+    input: ("text" | "image")[];
+}
+export declare const MODEL_CAPABILITIES: Record<string, ModelCapabilities>;
 export declare const PROVIDERS: Record<string, {
     env: string;
     label: string;

package/dist/types.js

@@ -1,3 +1,23 @@
+export const MODEL_CAPABILITIES = {
+    // Anthropic
+    "claude-sonnet-4-20250514": { contextWindow: 200000, maxTokens: 16384, reasoning: true, input: ["text", "image"] },
+    "claude-opus-4-0520": { contextWindow: 200000, maxTokens: 16384, reasoning: true, input: ["text", "image"] },
+    // OpenAI
+    "gpt-4o": { contextWindow: 128000, maxTokens: 16384, reasoning: false, input: ["text", "image"] },
+    "o3-mini": { contextWindow: 128000, maxTokens: 65536, reasoning: true, input: ["text"] },
+    // Google
+    "gemini-2.5-pro": { contextWindow: 1048576, maxTokens: 65536, reasoning: true, input: ["text", "image"] },
+    "gemini-2.5-flash": { contextWindow: 1048576, maxTokens: 65536, reasoning: true, input: ["text", "image"] },
+    // Groq
+    "llama-3.3-70b-versatile": { contextWindow: 128000, maxTokens: 32768, reasoning: false, input: ["text"] },
+    // OpenRouter
+    "anthropic/claude-sonnet-4": { contextWindow: 200000, maxTokens: 16384, reasoning: true, input: ["text", "image"] },
+    "openai/gpt-4o": { contextWindow: 128000, maxTokens: 16384, reasoning: false, input: ["text", "image"] },
+    // xAI
+    "grok-3": { contextWindow: 131072, maxTokens: 16384, reasoning: false, input: ["text", "image"] },
+    // Mistral
+    "mistral-large-latest": { contextWindow: 128000, maxTokens: 8192, reasoning: false, input: ["text"] },
+};
 export const PROVIDERS = {
     anthropic: { env: "ANTHROPIC_API_KEY", label: "Anthropic (Claude)", models: ["claude-sonnet-4-20250514", "claude-opus-4-0520"] },
     openai: { env: "OPENAI_API_KEY", label: "OpenAI (GPT)", models: ["gpt-4o", "o3-mini"] },

package/dist/utils/install.js

@@ -3,7 +3,7 @@ import { join, dirname } from "node:path";
 import { fileURLToPath } from "node:url";
 import { homedir } from "node:os";
 import { execSync } from "node:child_process";
-import { KEYBINDING_SCHEMES, PROVIDERS } from "../types.js";
+import { KEYBINDING_SCHEMES, MODEL_CAPABILITIES, PROVIDERS } from "../types.js";
 const PKG_ROOT = join(dirname(fileURLToPath(import.meta.url)), "..", "..");
 function ensureDir(dir) {
     mkdirSync(dir, { recursive: true });
@@ -35,13 +35,18 @@ export function applyConfig(config) {
     // 2. settings.json
     const primary = config.providers[0];
     const providerInfo = primary ? PROVIDERS[primary.name] : undefined;
+    const compactThreshold = config.compactThreshold ?? 0.75;
+    const primaryModel = primary?.defaultModel ?? providerInfo?.models[0];
+    const primaryCaps = primaryModel ? MODEL_CAPABILITIES[primaryModel] : undefined;
+    const contextWindow = primary?.contextWindow ?? primaryCaps?.contextWindow ?? 128000;
+    const reserveTokens = Math.round(contextWindow * (1 - compactThreshold));
     const settings = {
         defaultProvider: primary?.name,
-        defaultModel: primary?.defaultModel ?? providerInfo?.models[0],
+        defaultModel: primaryModel,
         defaultThinkingLevel: config.thinking,
         theme: config.theme,
         enableSkillCommands: true,
-        compaction: { enabled: true, reserveTokens: 16384, keepRecentTokens: 20000 },
+        compaction: { enabled: true, reserveTokens, keepRecentTokens: 20000 },
         retry: { enabled: true, maxRetries: 3 },
     };
     if (config.providers.length > 1) {
@@ -51,19 +56,41 @@ export function applyConfig(config) {
         });
     }
     writeFileSync(join(agentDir, "settings.json"), JSON.stringify(settings, null, 2));
-    // 3. keybindings.json
+    // 3. models.json (custom endpoints / providers)
+    const customProviders = config.providers.filter(p => p.baseUrl);
+    if (customProviders.length > 0) {
+        const models = {};
+        for (const cp of customProviders) {
+            const caps = cp.defaultModel ? MODEL_CAPABILITIES[cp.defaultModel] : undefined;
+            models[cp.name] = {
+                baseUrl: cp.baseUrl,
+                apiKey: cp.apiKey === "none" ? undefined : cp.apiKey,
+                api: "openai-completions",
+                models: cp.defaultModel ? [{
+                        id: cp.defaultModel,
+                        name: cp.defaultModel,
+                        reasoning: cp.reasoning ?? caps?.reasoning ?? false,
+                        input: cp.multimodal ? ["text", "image"] : (caps?.input ?? ["text"]),
+                        contextWindow: cp.contextWindow ?? caps?.contextWindow ?? 128000,
+                        maxTokens: cp.maxTokens ?? caps?.maxTokens ?? 8192,
+                    }] : [],
+            };
+        }
+        writeFileSync(join(agentDir, "models.json"), JSON.stringify(models, null, 2));
+    }
+    // 4. keybindings.json
     const kb = KEYBINDING_SCHEMES[config.keybindings];
     if (kb && Object.keys(kb).length > 0) {
         writeFileSync(join(agentDir, "keybindings.json"), JSON.stringify(kb, null, 2));
     }
-    // 4. AGENTS.md
+    // 5. AGENTS.md
     const agentsSrc = join(PKG_ROOT, "pi-package", "agents", `${config.agents}.md`);
     try {
         const content = readFileSync(agentsSrc, "utf8");
         writeFileSync(join(agentDir, "AGENTS.md"), content);
     }
     catch { /* template not found, skip */ }
-    // 5. Copy extensions (single file .ts or directory with index.ts)
+    // 6. Copy extensions (single file .ts or directory with index.ts)
     const extDir = join(agentDir, "extensions");
     ensureDir(extDir);
     for (const ext of config.extensions) {
@@ -79,7 +106,7 @@ export function applyConfig(config) {
             catch { /* skip */ }
         }
     }
-    // 6. Copy prompts
+    // 7. Copy prompts
     const promptDir = join(agentDir, "prompts");
     ensureDir(promptDir);
     for (const p of config.prompts) {
@@ -89,7 +116,7 @@ export function applyConfig(config) {
         }
         catch { /* skip */ }
     }
-    // 7. Copy skills
+    // 8. Copy skills
     const skillDir = join(agentDir, "skills");
     ensureDir(skillDir);
     for (const s of config.skills) {
@@ -101,7 +128,7 @@ export function applyConfig(config) {
         }
         catch { /* skip */ }
     }
-    // 8. Copy themes (only custom ones)
+    // 9. Copy themes (only custom ones)
     const themeDir = join(agentDir, "themes");
     ensureDir(themeDir);
     const themeSrc = join(PKG_ROOT, "pi-package", "themes", `${config.theme}.json`);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "oh-pi",
-  "version": "0.1.0",
+  "version": "0.1.2",
   "description": "One-click setup for pi-coding-agent. Like oh-my-zsh for pi.",
   "type": "module",
   "bin": {