Dev & Engineering swift-nativemacoscoding-agentclirustdeepseeklocal-inference

DeepTide - AI Coding Agent by DeepSeek

Native macOS app and cross-platform CLI coding agent tuned for DeepSeek, with BYOK support for any Anthropic-compatible endpoint.

FollowAgents review · FARS-2.1
Not recommended
44/ 100 5-point scale 2.2 / 5
1 2 3 4 5 6
1Trust8 / 29 · 1.4/5

Evidence shows permission modes (default, accept-edits, plan, bypass) and interactive tool approvals (GUI with colored diff preview), but no details on default permissions or enforcement. Data flow transparency is limited: session storage and config paths are mentioned, but not how data is transmitted or stored. Sensitive data handling: privacy reminders and vulnerability reporting guidance exist, but no details on API key storage. Dependency security: dependencies are pinned, but no vulnerability scanning or audit evidence. External effects: tools include shell and web fetch, but no default restrictions. Rollback: not mentioned. Source attribution: clearly based on Zero CLI, but publisher identity unverified. Deductions: lack of implementation details, rollback missing.

2Reliability6 / 14 · 2.1/5

Self-consistency: README and Cargo.toml are consistent, but multiple flavors (macOS, CLI, Rust) may cause confusion. Dependency availability: dependencies are pinned, but no availability guarantees. Failure messages: doctor command and error handling exist, but no specific failure message examples. Deductions: insufficient evidence for dependency availability and failure messages.

3Adaptability10 / 18 · 2.8/5

Audience and scenarios: targets macOS users, CLI users, and Rust users; scenarios include interactive REPL, one-shot queries, GUI. Capability boundaries: lists 30+ tools and 25+ commands, but no explicit boundaries. Trigger precision: command and tool names are clear, but no trigger condition details. Environment fit: supports multiple platforms, but macOS requires 15+, Rust version requires specific build. Deductions: insufficient detail on trigger precision and boundaries.

4Convention10 / 18 · 2.8/5

Information architecture: clear documentation structure with README, docs, samples, skills. Install notes: multiple installation methods provided. Naming stability: command names are stable, but multiple flavors may confuse. Examples and FAQ: samples and skills exist, but no FAQ. Known limitations: local model limitations mentioned, but not comprehensive. License: MIT, with LICENSE and NOTICE. Versioning and changelog: version numbers exist, but no changelog. Maintenance responsibility: issue templates and security policy exist. Deductions: missing FAQ, changelog, and comprehensive limitations.

5Effectiveness7 / 13 · 2.7/5

Output usability: provides multiple output formats (streaming, markdown, tool-call cards). Marginal value: offers local inference and multiple integrations, but value depends on user needs. Cost-benefit: no pricing or performance data. Deductions: insufficient evidence for cost-benefit.

6Verifiability3 / 8 · 1.9/5

Claim traceability: README claims features, but no test evidence. Cross-source corroboration: test vector script exists, but no independent verification. Fact-inference separation: documentation distinguishes facts and inferences, but not clearly. Deductions: lack of test results and independent verification.

Evidence confidence: Low Reviewed Aug 13, 2026 Reviewed revision 4604b1757131
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Safety controls not found in source: rollback or recovery path
Before you use it
  • Publisher identity is unverified; the curl | sh install script poses supply-chain risk.
  • Default permission mode is unclear, potentially allowing sensitive tool operations.
  • API key storage is not documented, posing leakage risk.
  • Dependencies are not vulnerability-scanned, risking known vulnerabilities.
  • Rollback mechanism is absent; erroneous operations may be unrecoverable.
Review evidence [1][2][3][4][5][6][7][8]
See the full review method →

What does this agent do, and when should you use it?

DeepTide is an AI coding agent built by DeepSeek, offered in multiple flavors: a native macOS app (Swift), a cross-platform CLI (npm package backed by Bun), and a Rust port with a desktop GUI. The CLI defaults to the DeepSeek API but can also drive any Anthropic-protocol-compatible endpoint via BYOK (e.g., Zhipu GLM, Volcengine, Qwen, Moonshot). The agent features a multi-turn agentic loop, 30+ built-in tools, 25+ slash commands, permission modes, a hooks engine, a memory system, and sub-agents. Besides the CLI and native app, the repo hosts an open-source local inference runtime under native/, including the ds4 inference engine and dsgo gateway. The CLI configuration is stored in ~/.config/tide/settings.json, with project-level overrides via .deeptide/settings.json. The desktop GUI shares the CLI's config, session history, and tool set.

DeepTide is an agentic coding assistant: the model plans, calls tools, observes results, and adapts. It talks to the DeepSeek API by default, or any Anthropic-compatible endpoint via --base-url and --api-key. Key commands include: tide login (save API key), tide auth login (Paean OAuth), tide to launch interactive REPL, tide -p for one-shot queries, tide doctor for diagnostics. It provides 30+ tools including file I/O, shell, WebFetch, MCP servers, sub-agents, and scheduling. It supports 25+ slash commands like /status, /cost, /diff. Permission modes include default, accept-edits, plan, and bypass. A hooks engine allows pre/post tool, user-prompt, session, and compaction hooks. The memory system persists project memory across sessions. The Rust flavor includes a desktop GUI (deeptide-rs --gui or deeptide-gui) with streaming chat, markdown rendering, reasoning display, tool-call cards, interactive approvals, a session sidebar, and a cost/usage bar. The local inference runtime consists of ds4 (DeepSeek V4 Flash Metal engine) and dsgo (OpenAI/Anthropic-compatible gateway).

  1. macOS users who want a tuned native Swift experience via the curl installer and use the tide command for interactive coding.
  2. Linux/Windows developers who install deeptide globally via Bun or npm and use the tide CLI for their workflow.
  3. Teams that want to connect DeepTide to non-DeepSeek providers like Zhipu GLM or self-hosted gateways using BYOK with --base-url and --api-key.
  4. Headless CI or low-spec laptop users who prefer the Rust binary (deeptide-rs) and optionally launch the desktop GUI with --gui.
  5. Researchers or developers who want to integrate DeepSeek V4 Flash local inference into their own workflows using the ds4 and dsgo components from native/.
  6. Organizations that need consistent configuration across CLI and GUI, sharing ~/.config/tide/settings.json and project .deeptide/settings.json.

What are this agent's strengths and limitations?

Pros
  • Offers multiple flavors (native macOS, CLI, Rust) to suit different user preferences and platforms.
  • Defaults to DeepSeek API but supports BYOK for any Anthropic-compatible endpoint, including Zhipu GLM, Moonshot, etc.
  • Rust desktop GUI shares configuration, sessions, and tool set with the CLI, eliminating duplicate setup.
  • Includes a local inference runtime (ds4 and dsgo) for offline or self-hosted scenarios.
  • Rich toolset and slash commands with MCP support, sub-agents, permission modes, and hooks engine.
  • MIT-licensed open-source code.
Limitations
  • CLI requires the Bun runtime even when installed via npm, adding an environmental dependency.
  • Local inference runtime ds4 has a context limit (64k) and higher-fidelity Q4_K build needs 192GB+ RAM.
  • macOS native app requires macOS 15+, limiting older systems.
  • CLI core is based on the third-party zero-cli project, introducing upstream dependencies.
  • Compatibility with Anthropic protocol may vary by provider (e.g., GLM may not be fully compliant).

How do you install or deploy this agent?

macOS (recommended): curl -fsSL https://deeptide.sh/install.sh | sh. Linux/Windows (CLI): install Bun (https://bun.com/) first, then run bun add -g deeptide or npm install -g deeptide or pnpm add -g deeptide. Rust CLI: npm install -g deeptide-rs. Build from source: clone the repo, for the macOS native app inspect the Swift source; for the Rust GUI run cargo build -p deeptide-gui --release.

How do you use this agent?

After installation, run tide auth login (or tide login to save a DeepSeek API key), tide to launch the REPL, tide -p "explain this repo" for a one-shot query, tide doctor for diagnostics. For BYOK: export DEEPSEEK_API_KEY="sk-..." or use --base-url and --api-key. Rust GUI: run deeptide-rs --gui or deeptide-gui. Configure API key before launch via export DEEPTIDE_API_KEY or settings.json.

How does this agent compare with similar options?

Compared to generic AI assistants like GitHub Copilot or Cursor, DeepTide is specifically designed for DeepSeek and offers multiple delivery modes (native, CLI, Rust). Versus OpenAI Codex or Claude Code, DeepTide supports a broader set of providers via BYOK and optional local inference.

FAQ

Is DeepTide free to use?
DeepTide itself is MIT-licensed open source, but using its features typically requires a DeepSeek API key or a BYOK provider key, which may incur costs.
Can I use the macOS native app on Windows?
No, the native app is for macOS 15+ only. Windows users should use the npm package deeptide (Bun-based) or the Rust CLI deeptide-rs.
How does DeepTide handle security and permissions?
It supports permission modes (default, accept-edits, plan, bypass) and interactive tool approvals (with colored diff previews in the GUI). It also advises users to redact sensitive data before pasting logs.
Can I run DeepSeek models locally?
Yes, the native/ directory contains the ds4 inference engine and dsgo gateway that can be built for local runtime. However, be aware of context limits (64k) and hardware requirements.

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related agents