Managing Your AI Environment

📝 Course Notes

Key takeaways from this lesson:

---

What You'll Learn

• 👀 View available models without entering the TUI
• 💰 Track token usage and identify which models cost the most
• 🔑 See which accounts you're logged into and log out when needed
• 🔄 Know how to refresh cache when configuration doesn't take effect

---

1. View Available Models

When you've configured multiple providers (like using both DeepSeek and Claude), or added a new key, you might ask: "What models can I actually use right now?"

Terminal Command: opencode models

Run directly in terminal:

opencode models

You'll see a list like this:

opencode/glm-4.7-free
anthropic/claude-3-5-sonnet-20241022
google/gemini-2.0-flash
ollama/deepseek-r1:7b
zhipuai-coding-plan/glm-4.7
...

Each line is a model ID (format: provider/model-name). You can copy this ID and specify it at startup:

# For example, start with Zhipu GLM-4.7
opencode --model zhipuai-coding-plan/glm-4.7

Advanced Tip 1: Filter by Provider

List too long? You can specify the provider name directly:

# Only show Anthropic models
opencode models anthropic

# Only show DeepSeek models (if configured)
opencode models deepseek

Much cleaner - only shows the models you care about.

Advanced Tip 2: View Pricing

Want to know the specific pricing for a model? Add the --verbose flag:

opencode models --verbose

The output will include detailed metadata, including inputCost and outputCost:

zhipuai-coding-plan/glm-4.7
{
  "id": "zhipuai-coding-plan/glm-4.7",
  "name": "GLM 4.7",
  "provider": "zhipuai-coding-plan",
  "inputCost": 0,    // $0!
  "outputCost": 0    // $0!
}

Advanced Tip 3: Refresh Cache

💡 Why don't I see the model after configuring the key?

OpenCode caches the model list. If you just added a new configuration in opencode.json, or just got access to a new model (e.g., downloaded a new Ollama model), the list may not update immediately.

In this case, you need to force refresh:

opencode models --refresh

After seeing the Models cache refreshed message, run opencode models again to see the new models.

Key Step: Set Your Favorite Model as Default

Found the model ID you want to use (e.g., zhipuai-coding-plan/glm-4.7), you don't want to type it every time you start, right?

Open your configuration file (usually at ~/.config/opencode/opencode.json) and add it:

{
  "model": "zhipuai-coding-plan/glm-4.7"
}

Now just type opencode and it will start with your favorite model.

---

2. Calculate Costs: View Usage Statistics

"How much did I spend coding this month?" "Which do I use more, Claude or GPT-4?"

OpenCode has a built-in statistics panel.

Terminal Command: opencode stats

Run:

opencode stats

You'll see a detailed dashboard. Here's real data after one month of use:

┌────────────────────────────────────────────────────────┐
│                       OVERVIEW                         │
├────────────────────────────────────────────────────────┤
│Sessions                                           948 │
│Messages                                        30,575 │
│Days                                                29 │
└────────────────────────────────────────────────────────┘

┌────────────────────────────────────────────────────────┐
│                    COST & TOKENS                       │
├────────────────────────────────────────────────────────┤
│Total Cost                                    $1232.56 │
│Avg Cost/Day                                    $42.50 │
│Input                                           530.6M │
│Output                                            9.9M │
│Cache Read                                      703.0M │
└────────────────────────────────────────────────────────┘

┌────────────────────────────────────────────────────────┐
│                      TOOL USAGE                        │
├────────────────────────────────────────────────────────┤
│ read               ████████████████████ 7270 (34.3%)   │
│ bash               ███████████          4074 (19.2%)   │
│ edit               ████████             3007 (14.2%)   │
└────────────────────────────────────────────────────────┘

Key Metrics Explained

For beginners, focus on these numbers:

1. Total Cost: Estimated cost in USD. OpenCode calculates based on each provider's public pricing.

2. Input vs Output:

   • Input: Content you send to AI + file content AI reads. Large volume but cheap.
   • Output: Code and responses AI writes. Small volume but expensive.

3. Cache Read (Context Cache): 🔥 Key money-saving metric!

   • This is money OpenCode saved for you.
   • When you have continuous conversations, repeated file content gets cached.
   • 703.0M means 700 million tokens were read from cache (usually 1/10 of input cost), significantly reducing usage costs.

4. Tool Usage:

   • Shows which tools AI uses most.
   • read (reading files) usually ranks first, showing AI is working to understand your code.
   • bash (running commands) and edit (modifying code) are also primary tools.

Advanced: Project-Specific Statistics (Requires Git)

☕ Note for non-developers

If you don't know Git, or think "mixed billing is fine", you can skip this section! This section is for advanced users who need to clearly separate "company project" and "personal project" bills.

By default, opencode stats shows your total bill across all projects.

If you're in a Git project and want to see only that project's costs, add the --project "" flag:

# Only see statistics for current project
opencode stats --project ""

Real case comparison (running in the opencode source repository):

• Running opencode stats (total): shows $1232.
• Running opencode stats --project "" (current Git project): shows $923.

This means about $300 was spent on other projects (or Global area).

💡 Core Logic: What counts as a "project"?

OpenCode identifies projects by the Git repository's first commit record.

• Project (separate billing): Git repository with at least one commit.
• Global (global billing): Everything else (regular folders, empty Git repos).

⚠️ Pitfall Guide: If you run opencode stats --project "" in a regular folder, the amount shown is the sum of all Global operations on this machine (e.g., usage in Downloads, Desktop, and all non-Git directories combined), not just the current folder.

Want to enable separate billing? Just three steps (no GitHub connection needed, local repo is fine):

1. git init
2. git add .
3. git commit -m "init" 👈 This step is crucial!

💡 Hidden Feature: Automatic Bill "Migration"

You might worry: "I've been coding in this folder for days, spent quite a few tokens, but never used Git. If I initialize a repo now, will the previous bills be lost?"

OpenCode is smarter than you think: It automatically transfers!

When you successfully create a Git project (and commit) in a folder, OpenCode automatically scans the "global ledger" and transfers all historical sessions belonging to this folder path to the new project.

So, use it confidently, no need to worry about whether to "activate membership first" or "buy things first".

---

See Who's the "Token Eater"

Want to know which model cost you the most? Run:

# Show top 5 most expensive models
opencode stats --models 5

# Show detailed list of all models
opencode stats --models

This will list the Top 5 detailed bills:

┌────────────────────────────────────────────────────────┐
│                      MODEL USAGE                       │
├────────────────────────────────────────────────────────┤
│ anthropic/claude-opus-4-5-20251101                     │
│  Messages                                       12,548 │
│  Input Tokens                                   241.0M │
│  Cost                                       $1232.5613 │
├────────────────────────────────────────────────────────┤
│ zhipuai-coding-plan/glm-4.7                            │
│  Messages                                        3,023 │
│  Input Tokens                                    68.9M │
│  Cost                                          $0.0000 │
├────────────────────────────────────────────────────────┤

The difference is clear at a glance:

• Claude Opus: Pay-per-use, powerful but really expensive, burned $1232.
• Zhipu GLM-4.7: Shows $0.0000, because I purchased a Coding Plan monthly subscription.

💡 Money-Saving Strategy

This is the purpose of OpenCode's statistics feature—it tells you where your money went. If you find a model with huge usage and high cost, consider switching to a subscription-based model (like Zhipu) as your main model, and save the expensive pay-per-use models for the hardest problems.

So, checking this statistics regularly can help you optimize your model combination and save real money.

Advanced Tip: View Recent Usage Only

By default, stats shows cumulative historical totals. But you might care more: "How much did I spend today?" or "How much did that new model cost me this week?"

Use the --days flag to specify the number of days:

# Only show statistics for the past 24 hours
opencode stats --days 1

# View the last 7 days
opencode stats --days 7

🔥 Power Combo:

Combine "time limit" and "model statistics" to precisely identify recent "big spenders":

# See who's spending money today (list model consumption for past 24 hours)
opencode stats --days 1 --models

---

3. Check Credentials: Manage Identity

Over time, you may have tried many models and configured many keys. What credentials are stored in your OpenCode now?

View Credential List

opencode auth list

You'll see a tree diagram like this:

Credentials ~/.local/share/opencode/auth.json
┌
●  Zhipu AI Coding Plan  api
│
●  Google  oauth
│
●  OpenAI  oauth
│
└  3 credentials

This command tells you:

1. Authentication Method:
   • api: Added by entering sk-... key.
   • oauth: Authorized via web redirect (like Google, OpenAI).
2. Storage Location: Usually at ~/.local/share/opencode/auth.json.

   ⚠️ Security Note

   This file stores your keys in plain text JSON format. Make sure to protect your computer and this file, and never upload it to a public code repository.

Logout/Delete Credentials

If you want to delete an old key (e.g., key leaked or expired):

opencode auth logout

Then press ↑ ↓ to select the provider to delete, and press Enter to confirm.

---

4. Corresponding Operations in TUI

The above are operations when you're in the terminal (not inside OpenCode interface). If you're already in OpenCode, there are corresponding shortcut commands:

What You Want to Do	Terminal Command	TUI Slash Command
Switch model	opencode -m <id>	/models
Login account	opencode auth login	/connect
View status	opencode stats	/status (shows summary)
Exit program	Ctrl+C	/exit

Difference

The /models command in TUI not only shows models but also directly switches the current session's model.

---

Checklist ✅

• [ ] Run opencode models to confirm you see your configured models
• [ ] Run opencode stats to glance at your token usage (is it less than you imagined?)
• [ ] Run opencode auth list to clean up old credentials you don't need

---

Appendix: Source Code Reference

Click to expand and view source code locations

Last updated: 2026-01-19

Feature	File Path	Lines
View Models	src/cli/cmd/models.ts	Entire file
Statistics Logic	src/cli/cmd/stats.ts	107-301
Project ID Recognition	src/project/project.ts	50-170
Auth List	src/cli/cmd/auth.ts	170-215
Auth Storage	src/auth/index.ts	38-73

Key Logic Explained:

1. Project ID Generation (project.ts):

   • OpenCode looks up for .git directory.
   • If found, tries to get root commit hash as project ID.
   • If no .git or no commit, falls back to ID "global". This is why empty folders share the same global billing.

2. Bill Migration Mechanism (project.ts):

   • When OpenCode detects the current project ID is not global, it triggers migrateFromGlobal.
   • It iterates through all Global sessions, checking if session.directory matches the current workspace.
   • If it matches, automatically updates that session's projectID to the new project's ID, achieving "bill transfer".

3. Statistics Filtering (stats.ts):

   • By default (when projectFilter is undefined), all sessions are counted.
   • Only when --project "" is explicitly passed, it calls getCurrentProject() to get the current directory ID and filters.

4. Credential Storage (auth/index.ts):

   • Credentials are stored in ~/.local/share/opencode/auth.json.
   • Code uses JSON.stringify directly for writing, unencrypted.
   • Only file permission 0o600 (current user read/write only) is set as a security measure.

---

With your environment managed, Phase 2 learning is complete.

Next, we enter Phase 3: Efficient Workflows.

In the next lesson, we'll learn 3.1 Plan vs Build.

You'll learn:

• What is Plan (thinking) mode
• What is Build (execution) mode
• How to switch between these modes for smarter AI work

---

Next Lesson Preview

In the next lesson, we'll learn 2.6 Git Basics.

You'll learn:

• Use commits to record "returnable save points"
• Sync local repository to GitHub
• Why Git makes certain undo/rollback operations more reliable