What is AgentPM?
AgentPM is an evidence layer for coding-agent work. It captures local agent sessions, makes them searchable, and helps teams understand what happened before work becomes a pull request, ticket, or review.
// faq
Questions about AgentPM and coding-agent observability.
Direct answers for teams evaluating AgentPM as the evidence layer for Claude Code, Codex, Cursor, Grok, and agent-assisted software work.
// start here
Decide whether AgentPM fits your team.
Learn what AgentPM captures, how it differs from LLM observability, how it works with Claude Code, Codex, Cursor, and Grok, and where it fits beside GitHub, Jira, Linear, and code review.
// product basics
Product Basics
Who is AgentPM for?
AgentPM is for developers, engineering leads, product leaders, staff engineers, and technical teams already using coding agents to ship real software.
What problem does AgentPM solve?
AgentPM solves the visibility gap around agent-assisted development. It preserves prompts, commands, retries, decisions, files, and open loops that usually disappear into local chat history and terminal logs.
Is AgentPM another productivity dashboard?
No. AgentPM focuses on evidence, review, search, and learning from real agent sessions. The goal is to help teams understand agent-assisted work, not reduce developers to vanity metrics.
// coding agent observability
Coding Agent Observability
What is coding agent observability?
Coding agent observability is visibility into the work humans and coding agents do together: prompts, agent replies, commands, tool output, file edits, decisions, errors, and unfinished tasks.
How is AgentPM different from LLM observability?
LLM observability usually tracks model calls, traces, cost, latency, and evals. AgentPM tracks engineering work: what agents attempted, where they got stuck, what changed, and what evidence explains the session.
What is prompt-to-PR evidence?
Prompt-to-PR evidence is the session history that explains how a human request became code: prompts, plans, commands, tool output, edits, tests, decisions, and unresolved risks.
Why not just read the final pull request?
A pull request shows the final diff and review discussion. It usually does not show failed attempts, local command output, agent assumptions, retries, or earlier decisions that shaped the implementation.
// supported coding agents
Supported Coding Agents
Does AgentPM work with Claude Code?
AgentPM is designed around local coding-agent workflows including Claude Code. It captures local session evidence and routes it into the right organization.
Does AgentPM work with Codex?
Yes. AgentPM supports Codex-style local agent sessions where prompts, tool calls, commands, file edits, verification, and open loops all matter to the team.
Does AgentPM work with Cursor?
Yes. AgentPM supports Cursor session capture for local coding-agent workflows, including transcript evidence that helps teams inspect prompts, replies, edits, tool output, and follow-up work.
Does AgentPM work with Grok?
Yes. AgentPM supports Grok session and log JSONL evidence from local machines, then normalizes it into searchable conversations for review and team learning.
Does AgentPM change how coding agents run?
No. AgentPM adds a collector and evidence layer around the work. Developers continue using their coding agents in local repos.
Can AgentPM support other coding agents?
AgentPM is built for local coding-agent workflows and already supports Claude Code, Codex, Cursor, and Grok. The collector and parser model can expand to other agent-assisted development environments as teams need them.
// session capture
Session Capture
What does AgentPM collect?
AgentPM focuses on coding-agent session evidence: human prompts, agent replies, commands, tool output, file context, decisions, artifacts, risks, and summarized signals that help teams review and improve agent-assisted work.
Does AgentPM capture terminal output?
Yes, when terminal and command output are part of the coding-agent session record, AgentPM keeps that evidence with the surrounding conversation context.
Can AgentPM show what tests or commands the agent ran?
Yes. AgentPM keeps command execution and tool output with the session so reviewers can see what the agent attempted and where verification succeeded or failed.
What happens when a session is interrupted or unfinished?
AgentPM preserves the available session evidence and can surface open loops, errors, and unfinished work so teams know what still needs human attention.
// search and review
Search and Review
Can teams search old agent sessions?
Yes. AgentPM makes captured sessions searchable so teams can find prior decisions, commands, errors, implementation attempts, and reusable workflow patterns.
Can AgentPM help review agent-generated code?
AgentPM helps reviewers inspect the session context behind agent-assisted code, including what the agent was asked to do, what it tried, what it verified, and what it left unresolved.
What is an agent audit trail?
An agent audit trail is a durable record of agent-assisted work: the request, agent activity, commands, tool output, decisions, verification attempts, and open risks around a software change.
Does AgentPM prove that code is correct?
No. AgentPM does not prove correctness. It gives reviewers better evidence so they can focus human judgment, testing, and review on the right risks.
// engineering leadership
Engineering Leadership
How does AgentPM help engineering leads?
AgentPM helps leads see where agent-assisted work is happening, where agents and developers get stuck, which practices are working, and what guidance the team should improve next.
Can AgentPM show agent usage across repos or machines?
AgentPM is organized around teams, organizations, connected machines, directories, and sessions, so leaders can understand adoption patterns without asking every developer to reconstruct their day.
How does AgentPM help with agent rollout?
AgentPM gives teams evidence from real sessions, making it easier to spot setup problems, repeated workflow friction, training needs, and successful patterns worth standardizing.
Is AgentPM for monitoring developers?
AgentPM is designed as an engineering evidence layer, not a surveillance product. Its value is helping teams review work, reduce risk, and improve agent workflows from real evidence.
// setup, security, and fit
Setup, Security, and Fit
Does AgentPM replace GitHub, Jira, or Linear?
No. AgentPM captures the work that happens before those systems have a clean artifact. It complements code hosts, issue trackers, and review tools with searchable session evidence.
How does setup work?
A team creates or opens an AgentPM organization, installs the local collector on a developer machine, verifies the agent is sending, and then reviews incoming sessions in Overview, Notebook, Insights, and Coaching views.
Can AgentPM help with sensitive work?
AgentPM is built for engineering evidence, so teams should treat captured sessions as sensitive engineering data. Access, rollout, and retention choices should match the team's security expectations.
When should a team adopt AgentPM?
AgentPM is most useful once a team is using coding agents for real work and needs shared visibility into what agents attempted, changed, missed, or left unresolved.
// related use cases
Explore the workflows behind the answers
Category
Coding Agent Observability
See the coding-agent work that happens before a pull request, ticket, or review: prompts, commands, retries, files, decisions, and unresolved loops.
Claude Code
Claude Code Team Visibility
Give teams a shared view of Claude Code sessions without asking developers to manually summarize prompts, retries, commands, and decisions.
Codex
Codex Session Tracking
Track Codex work from prompt to local commands, file edits, test runs, open loops, and review-ready evidence.