Documentation
Concepts
The ideas behind AgentPM: conversations, evidence, insights, search, knowledge graph, and compounded organizational knowledge.
The AgentPM model
AgentPM starts with captured coding-agent sessions and turns them into searchable, reviewable, team-owned work history.
Conversations
A conversation is the normalized record of a coding-agent work session. It includes prompts, assistant turns, tool calls, command output, file references, metadata, and derived summaries.
OpenEvidence
Evidence is the inspectable trail behind agent-assisted work: commands, files, tests, branches, decisions, risks, and exact turns. AgentPM keeps the record reviewable after the chat window is gone.
OpenInsights
Insights turn long sessions into typed keyframes such as decisions, outcomes, risks, artifacts, configuration changes, advice, and unresolved work.
OpenDaily Digest
Daily Digest summarizes a day of agent-assisted engineering work across projects, conversations, risks, and follow-ups so teams can operate from shared context.
OpenOrganizations, projects, and sessions
Organizations define the team boundary. Projects group work by repository or working directory. Sessions are the captured agent runs that feed conversations, insights, search, and digests.
OpenSearch
AgentPM search lets teams find conversations by natural language, facets, file paths, commands, projects, agents, dates, and git activity.
OpenKnowledge Graph
The knowledge graph connects conversations to projects, files, agents, commands, skills, MCP calls, branches, PRs, decisions, risks, and recurring patterns.
OpenCompounded organizational knowledge
Every captured session can improve the next session. AgentPM helps teams turn repeated work into durable memory, project guidance, skills, review practices, and better agent workflows.
Open