OPEN SOURCE · MIT · LOCAL-FIRST

Your AI agent shouldn't be a black box

Pick any modelSee every callKnow the cost

Get started Explore the product

DeepAA / Dashboard

DeepAADashboardUTC+8

CodexLaunchClaude CodeLaunchOpenCodeLaunchDeepSeek HarnessLaunchZCodeLaunch

bigmodelPlan

Available quota68%

5 hours / 1 week

catapiPay-as-you-go

Account balance¥ 9.45

Pay-as-you-go

chatgptSubscription

Available quota100%

Sync failed

auto-codePay-as-you-go

Account balance$ 1005.33 (≈ ¥62.83)

Pay-as-you-go

opencodePlan

Available quota100%

5 hours / 1 week / Month

Total requests1,676

98.3% success · 11 auxiliary requests excluded

Total tokens451M

Input, output, and cache usage in one view

Combined cost¥55.01

Usage charges + estimated plan cost

Usage ¥1.62Plan estimate ¥53.39

Token usage over time

Cache hit rate 95.9%

24-hour usage · millions of tokens

Token cost ranking

Estimated cost per million tokens

grok-4.7opencode¥ 8.89/M

gpt-5.6-solchatgpt¥ 0.41/M

gpt-6.1-solauto-code¥ 0.36/M

gpt-6.1-solchatgpt¥ 0.24/M

Usage by dimension

Model

glm-5.3356M

glm-5.3-flash50M

gpt-6.1-sol38M

Provider

bigmodel406M

catapi28M

chatgpt8.03M

Agent

ZCode409M

Codex39M

DeepSeek Harness3.14M

Top three in each dimension · full analytics in your local console

Your AI workflow, in full view

Launch agents, balances, usage, and cost in one console.

Sample snapshot · 2026-10-09 · UTC+8 · Static product sample. Launching, syncing, and configuration happen in your local console.

Explore the full product UI

WORKS WITH THE TOOLS YOU ALREADY USE

CodexClaude CodeOpenCodeDeepSeek HarnessZCode

OpenAI Responses / Chat Completions · Anthropic Messages

Official APIs, compatible relay services, and local models. Connected.

CALL TRACING

Beyond the result. Evidence at every step.

Follow a task from session to step. Connect model requests, tool calls, and responses, then inspect the input, output, and usage that matter.

Explore call tracing

DeepAA session tracing showing task hierarchy, tool steps, and request evidence — Sample product interfaceView the full screenshot ↗View the full screenshot ↗

Session → Thread → Turn → Step

01Follow tasks, sub-agents, and steps in one chain

02Inspect prompts, tool arguments, and results on demand

03Connect each step to its tokens, latency, and cost

ONE CONSOLE FOR YOUR AI WORKFLOW

See clearly. Work reliably. Understand cost.

USAGE & COST

Know where every token goes.

Inspect per-request costs and formulas, from token rates, caching, and credential multipliers to plan usage rules. Compare totals by model, provider, and agent.

Understand cost reporting

Combined sample cost¥55.01

Usage charges¥1.62

Estimated plan cost¥53.39

Prices, multipliers, and exchange rates are frozen per call for traceable records.

UNIFIED ACCESS

Your models and providers. One place to manage them.

Check provider balances and plan quotas, sync managed agent settings, and use compatible models from multiple providers in the same agent.

CodexClaude CodeOpenCode

DeepAALocal gatewayOfficial / relay / local models

Read the setup guide

REQUEST EVIDENCE

Keep the details. Find the problem.

Inspect requests, responses, and streaming diagnostics. Copy curl commands and export content within an explicit scope.

PromptinputTool Callcall → resultResponseoutputSSEstream

Explore evidence & export

RELIABLE CALLS

When one channel fails, backups take over.

Configure ordered model-and-provider backups, with quick launch and model switching for everyday coding workflows.

Read about failover

Primary modelModel × provider

Backup modelModel × provider

Next backupModel × provider

LOCAL-FIRST, BY DESIGN

Your observability data. On your machine.

Captured requests, responses, and analytics stay on your machine, with no upload to a DeepAA cloud. No account required. Open code. Your data.

Read about data & privacy

Your machine

Coding agentDeepAA

Model provider

Local observability & analytics~/.deepaa

Model requests still go to the provider you choose. Observability records are not uploaded to DeepAA.

Local storage

Observability data is stored and analyzed locally

No account

Install and use your local console directly

MIT licensed

Open source, issues, and release history

GET STARTED

Make your next call a little clearer.

Choose an install method. One command opens your local console.

curl -fsSL https://deepaa.dev/install.sh | bash

The script checks or installs Node.js, then starts DeepAA.

Or ask your local agent to install it

One-Prompt Install via Your Local AgentRecommended

Copy the prompt below and paste it into Claude Code, Codex, Cursor, or any local agent — it will install and launch DeepAA for you.

Install and start DeepAA (a local-first AI agent gateway, observability and analytics tool) on this machine:
1. Check the Node.js version — 22.13+ is required; if missing or outdated, install the LTS release first;
2. Run npm install -g deepaa for a global install; on a slow network, add --registry=https://registry.npmmirror.com to this install only; do not change global npm settings;
3. Once installed it auto-launches and opens http://localhost:3210 (proxy on port 3211); if the browser did not open, run deepaa to start it;
4. Verify the Web UI is reachable and report the URL back to me.

BEFORE YOU INSTALL

A few things worth knowing

Read the full documentation

How does DeepAA work with my coding agent?

DeepAA is a local gateway and observability console. Keep using Codex, Claude Code, OpenCode, DeepSeek Harness, or ZCode while DeepAA helps connect models and inspect calls, usage, and cost.

Do I need to change agent source code or add an SDK?

No. Configure the local gateway for OpenAI / Anthropic compatible requests. Supported official-direct clients can also import local logs. On first use, explicitly select your agent, provider, model, and credentials in the console.

Where is my data stored?

Observability data is stored in ~/.deepaa on your machine, with no upload to a DeepAA cloud. Local services bind to 127.0.0.1 by default. Model requests still go to your chosen provider. Raw records can contain sensitive information; review them before sharing.

Are plan costs and usage charges reported the same way?

Usage charges use the price, multiplier, and exchange rate recorded at call time, with reconciliation adjustments where supported. Plans and subscriptions report estimated costs; unavailable estimates are marked as pending rather than presented as actual bills.

Which systems and model services are supported?

One-line installation supports macOS and Windows. The gateway accepts OpenAI Responses, Chat Completions, and Anthropic Messages compatible endpoints, including official APIs, compatible relays, and local models. Agent and model capabilities depend on your connection settings.