depthfirst's Dependency Firewall inspects every open-source package being pulled into a company, returns a verdict before install, and blocks malicious ones before they reach the developer, CI pipeline, or AI agent that requested them. The launch adds an on-device enforcement mode alongside the existing registry proxy, so security teams can push policies through MDM to laptops and agent runtimes without changing how engineers, sales, finance, or coding agents like Claude and Codex install packages. Approved packages pass through with negligible latency, suspicious ones are quarantined, and every verdict ships with the underlying evidence.
The timing tracks a specific shift in who installs software. The people pulling in dependencies now include business users running AI assistants and autonomous coding agents, not only security-conscious engineers , which expands the blast radius of install-time malware that can leak credentials, plant backdoors, or exfiltrate source code the moment a package is fetched. Rather than waiting for install time, depthfirst analyzes packages the moment they are published, running proprietary code and install-script analysis, runtime behavior detection, publisher and maintainer anomaly checks, transitive dependency mapping, and threat feed lookups. Teams can layer their own guardrails on top, including minimum package age, allowed dependency trees, and license policies, with overrides logged automatically.
depthfirst is a San Francisco security company founded in 2024 by Qasim Mithani, who led infrastructure and enterprise platform teams as Head of Infra at Databricks and held senior engineering roles at AWS, Daniele Perito, a Faire co-founder who was part of the founding team at Cash App and a postdoctoral security researcher at UC Berkeley, and Andrea Michi, previously a research engineering lead at Google DeepMind working on reinforcement learning for Gemini . Dependency Firewall is the newest addition to a product suite that already covers code scanning, supply chain reachability, secrets, and agentic pentesting, and it is available now for teams that want a single enforcement path across registries, package managers, and agent tooling.
Comments (9)
Sign in to join the discussion.
P
Priya Kalra23d ago
Dependency firewall is a great phrase until legal asks who owns the blocklist. Curious how you handle the inevitable false positive on a package someone shipped to prod at 2am.
T
Tomasz Wróbel23d ago
The launch video moves, but the on-device angle got buried under three logos in the first five seconds. Lead with the malicious-package block, not the architecture diagram.
C
Chidi Okafor23d ago
Poked at the docs looking for a CLI hook and found policy-as-code buried three pages deep. Promote that, devs will care more about it than the dashboard screenshots.
M
Mei L.23d ago
Every supply chain security tool promises 'without changing how employees work' and then ships a 40MB agent that fights with Rosetta. Prove me wrong and I'll actually try it.
R
Ravi S.23d ago
Naive question but what happens when the agent itself pulls a dep in the background? Do you sandbox the agent's package manager too or just the human's?
A
Annika Hvass23d ago
Tagline rewrite, on the house: 'Block bad packages before they block your Friday.' You can Venmo me later.
B
Burak Özdemir23d ago
Seat-based pricing on a security control that scales with number of registries feels like leaving money on the table. Or gouging customers. Depends which side of the invoice you're on.
S
Sofía Parra23d ago
Been shipping in the SCA space and this on-device piece is the part everyone waves at and nobody builds. Respect for actually doing it, now do the IDE plugin.
D
Deva K.23d ago
An IDE plugin is table stakes but the real win is blocking at the resolver, before lockfile writes. Curious if depthfirst hooks pnpm and uv or just npm and pip.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
depthfirst
The Operating System for Modern Security
The Operating System for Modern Security
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
depthfirst's Dependency Firewall inspects every open-source package being pulled into a company, returns a verdict before install, and blocks malicious ones before they reach the developer, CI pipeline, or AI agent that requested them. The launch adds an on-device enforcement mode alongside the existing registry proxy, so security teams can push policies through MDM to laptops and agent runtimes without changing how engineers, sales, finance, or coding agents like Claude and Codex install packages. Approved packages pass through with negligible latency, suspicious ones are quarantined, and every verdict ships with the underlying evidence. The timing tracks a specific shift in who installs software. The people pulling in dependencies now include business users running AI assistants and autonomous coding agents, not only security-conscious engineers , which expands the blast radius of install-time malware that can leak credentials, plant backdoors, or exfiltrate source code the moment a package is fetched. Rather than waiting for install time, depthfirst analyzes packages the moment they are published, running proprietary code and install-script analysis, runtime behavior detection, publisher and maintainer anomaly checks, transitive dependency mapping, and threat feed lookups. Teams can layer their own guardrails on top, including minimum package age, allowed dependency trees, and license policies, with overrides logged automatically. depthfirst is a San Francisco security company founded in 2024 by Qasim Mithani, who led infrastructure and enterprise platform teams as Head of Infra at Databricks and held senior engineering roles at AWS, Daniele Perito, a Faire co-founder who was part of the founding team at Cash App and a postdoctoral security researcher at UC Berkeley, and Andrea Michi, previously a research engineering lead at Google DeepMind working on reinforcement learning for Gemini . Dependency Firewall is the newest addition to a product suite that already covers code scanning, supply chain reachability, secrets, and agentic pentesting, and it is available now for teams that want a single enforcement path across registries, package managers, and agent tooling.
Comments (9)
Dependency firewall is a great phrase until legal asks who owns the blocklist. Curious how you handle the inevitable false positive on a package someone shipped to prod at 2am.
The launch video moves, but the on-device angle got buried under three logos in the first five seconds. Lead with the malicious-package block, not the architecture diagram.
Poked at the docs looking for a CLI hook and found policy-as-code buried three pages deep. Promote that, devs will care more about it than the dashboard screenshots.
Every supply chain security tool promises 'without changing how employees work' and then ships a 40MB agent that fights with Rosetta. Prove me wrong and I'll actually try it.
Naive question but what happens when the agent itself pulls a dep in the background? Do you sandbox the agent's package manager too or just the human's?
Tagline rewrite, on the house: 'Block bad packages before they block your Friday.' You can Venmo me later.
Seat-based pricing on a security control that scales with number of registries feels like leaving money on the table. Or gouging customers. Depends which side of the invoice you're on.
Been shipping in the SCA space and this on-device piece is the part everyone waves at and nobody builds. Respect for actually doing it, now do the IDE plugin.
An IDE plugin is table stakes but the real win is blocking at the resolver, before lockfile writes. Curious if depthfirst hooks pnpm and uv or just npm and pip.