Aseem Shrey
Aseem Shrey is an individual security engineer, ethical hacker, and solo developer/creator who builds security and privacy tooling. His principal AgentJail project is an open-source, offline policy guardrail for coding agents that evaluates tool calls before execution and combines Rego policy, OS-native sandboxing, network controls, credential boundaries, and auditability.
At a Glance
- Individual developers and security-conscious coding-agent users
- Engineering teams using Claude Code, Codex, Cursor, or MCP servers
- Security, platform, DevSecOps, and infrastructure teams governing AI-agent access
- Organizations needing local-first, auditable policy controls for agent actions
AI Tools by Aseem Shrey
(1)agentjail
Open Source Coding Agent Sandbox
Discussions
No discussions yet
Be the first to start a discussion about Aseem Shrey
Latest News
AgentJail v1.8.2 published on GitHub; maintenance release with managed core-policy refresh and security/release updates.
v1.8.1: clearer CLI help and consistent Codex approvals, including current Codex launch flags and retained reviewed execpolicy approvals.
v1.8.0: clearer approvals and faster cost reports; host commands require a reason and reports use a resumable local usage index.
v1.7.0: bounded authority visible to the agent; governed host-access discovery and exact session-bound runtime grants.
Products & Services
Open-source Apache-2.0 policy guardrails for coding agents. It intercepts tool calls at the PreToolUse boundary, evaluates local Rego/OPA rules, and returns allow/ask/deny decisions. Optional OS-native sandboxing uses Landlock on Linux and Seatbelt on macOS; network visibility uses an opt-in transparent tunnel; the project also provides MCP controls, credential brokering, session/audit views, cost analytics, and a local UI.
Market Position
AgentJail positions itself as a lightweight, open-source middle path between prompt/instruction files and coarse container or VM isolation: it makes deterministic Rego decisions at the tool boundary, while optionally adding OS-native sandboxing, network visibility, and audit controls. Its own comparison contrasts prompt-based controls, Docker/VM boundaries, and AgentJail; practical alternatives and competitors include Docker/VM sandboxes, built-in agent permission systems, and instruction-file guardrails such as CLAUDE.md, agents.md, and .cursorrules.
Leadership
Founders
Aseem Shrey
Security engineer and ethical hacker with 7+ years in cloud security, application security, and security automation. His published resume lists work at Scale AI, Rippling, Yahoo, GoJek, and Blinkit; his public profile also identifies him as founder/security architect of ShipSecAI and founder of AgentJail.
Executive Team
Aseem Shrey
Founder and Security Architect, AgentJail
Security engineer, ethical hacker, and security-tool builder; published background includes Scale AI, Rippling, Yahoo, GoJek, and Blinkit, plus founding ShipSecAI.
Founding Story
Shrey started AgentJail because coding agents can delete files, leak secrets, or make unsafe network and Git changes, while prompt-based guardrails and repeated permission prompts are unreliable or fatiguing. The initial vision was a small, open-source Apache-2.0 guardrail positioned at the tool boundary: deterministic local policy decisions before a dangerous call reaches the shell, with a practical alternative to all-or-nothing sandboxes.
Business Model
Revenue Model
AgentJail is distributed as open-source software under the Apache License 2.0, with installation from the public website and source code on GitHub. The reviewed public sources describe the CLI, local UI, and documentation but do not specify a paid subscription, API-usage charge, or commercial pricing model.
Target Markets
- Individual developers and security-conscious coding-agent users
- Engineering teams using Claude Code, Codex, Cursor, or MCP servers
- Security, platform, DevSecOps, and infrastructure teams governing AI-agent access
- Organizations needing local-first, auditable policy controls for agent actions
- Safely running Claude Code, Codex, or Cursor against local repositories
- Allowing agents to work with permission prompts reduced or disabled while retaining deterministic controls
- Preventing destructive shell commands, force-pushes, package publishing, credential reads, and environment-secret exfiltration
- Governing MCP server and tool access across developer workflows
- Sandboxing untrusted code and controlling filesystem, network, Git, and cloud/database credential access
- Auditing agent actions, sessions, model traffic, and local usage costs