AgentPM is a developer tool that automatically captures and indexes the complete work history of AI coding agents—conversations, commands, code changes, decisions, tests, and outcomes—into a searchable, source-linked record. The platform lets humans review what agents did and enables future agents to retrieve relevant context and continue from verified states instead of rebuilding understanding. AgentPM supports popular agents like Claude Code, Codex, Cursor, and Grok, with search features that work across team sessions and a CLI installer for macOS and Linux.
Funding
Funding not disclosed
Founders
Product
Problem
Teams using AI coding agents lack visibility into what those agents actually did, why they made specific decisions, and what state they left the work in. Useful information from past agent sessions—conversations, commands, tests, and outcomes—lives in forgotten chat scrollback, forcing humans to piece together context and future agents to repeat failed approaches or reopen settled questions.
Solution
AgentPM automatically captures conversations, commands, code changes, decisions, tests, and outcomes from coding agent sessions and stores them in one searchable history. It distills each session into one-line summaries and typed keyframes—decisions, outcomes, risks, artifacts, config, and state changes—each linked to the exact turn that produced it. Humans can search and review the complete record, while future agents retrieve the relevant conversations, decisions, and outcomes they need before planning, continuing from the latest verified outcome. The tool runs as a lightweight daemon via a single shell-install command, works with the coding agents teams already use, and supports multiple monitoring scopes from a single home directory to multiple machines.
Target Audience
Primary users are developers who need to recover context and keep work moving across agent sessions, engineering leaders who want to see patterns, outcomes, and bottlenecks, and security/governance teams that need auditability and control over agentic software development.
Features
- Automatic capture of conversations, commands, code changes, decisions, tests, and outcomes without requiring a new workflow
- Typed keyframes for filtering by decisions, outcomes, risks, artifacts, config, and state changes, each linking back to the exact source turn
- Full-text search across all team conversations, supporting cross-agent results from Claude Code, Codex, Cursor, Grok, OpenCode, Windsurf, and other supported history sources
- One-line session summaries that surface repeated friction, useful advice, and improvement opportunities
- Source-linked work history and personal notebooks for context recovery
- Installer script with configurable monitoring scope (single home, multiple homes, or specific project directories), and background service support via systemd or launchd
- SSO/SAML, RBAC, audit logs, retention controls, data residency, and private or VPC deployment options for team governance