dzhng/deep-research

Selection

Select when

• You want a standalone interactive deep-research workflow rather than a Claude Code extension.
• You need explicit controls for research breadth and recursive depth.
• You want Firecrawl-based search and extraction combined with LLM synthesis.
• You want a Markdown report written to the working directory.
• You want a compact TypeScript codebase intended to be easy to understand and extend.
• You can supply Firecrawl access and either hosted-model credentials or a compatible local endpoint.
• You want configurable concurrency for paid or self-hosted Firecrawl capacity.

Boundaries

Avoid when

• You need native Claude Code skills, hooks, slash commands, MCP integration, or context injection.
• You require a tested, production-ready research system with documented reliability guarantees.
• You cannot send queries or retrieved content to configured external providers.
• You need fully offline retrieval and inference.
• You need a graphical desktop or browser interface.
• You need built-in citation verification or independently evaluated research accuracy.
• You need a repository with a working automated test suite.
• You require an unambiguous license declaration across repository files.

Strengths

Capabilities

Risk profile

Risks and limitations

• Uncertainty: only README.md and package.json were supplied, so the agent implementation, prompts, recursion controls, source handling, Docker configuration, and API route were not directly inspected.
• Uncertainty: supplied metadata contains update and push dates in 2026 relative to this analysis context, so current activity and repository state cannot be independently reconciled.
• Uncertainty: no first-party evidence demonstrates any integration with Claude Code, so classification as a standalone companion or alternative is inferred from its own CLI workflow.
• The package is version 0.0.1 and configures no tests, which suggests early maturity but does not by itself prove instability.
• Research quality, factual accuracy, source relevance, and citation completeness were not benchmarked or independently validated.
• The README requires Firecrawl and OpenAI keys, while also documenting local and alternative model paths; exact minimum credential requirements for every configuration remain uncertain.
• Free Firecrawl usage may encounter rate limits and require concurrency to be reduced, making runs slower.
• Recursive breadth, depth, and concurrency can increase API consumption, latency, and cost.
• The system writes report.md or answer.md into the working directory, so existing-file behavior should be verified before use.
• The metadata and README identify the license as MIT, but package.json declares ISC, creating an unresolved licensing inconsistency.
• No security policy, threat model, privacy statement, telemetry disclosure, or independent audit was supplied.
• The undocumented api script and Express surface cannot be assessed for authentication, exposure, or production safety from the supplied files.

The workflow consumes service credentials from .env.local and sends research queries and retrieved content to Firecrawl and the configured model endpoint. A self-hosted Firecrawl base URL and local OpenAI-compatible model endpoint are documented alternatives, but the supplied files do not establish secret-storage protections, network boundaries, input hardening, API authentication, dependency review, or sandboxing. The repository has no configured tests and no supplied security policy or audit.