Developer · Free
Get started today.
- Capture and summarize your agent sessions
- Search conversations, commands, decisions, and outcomes
- Keep personal notebooks and source-linked work history
PUBLIC BETA
Remember what your coding agents actually did.
AgentPM gives humans and agents access to the conversations, decisions, and actions behind every session—so they can learn, improve, and continue work across context, models, and frameworks.
$ curl -fsSL https://agentpm.dev/install.sh | bash
~/src/NewTradingThing — claudeContinue work
User
proof: agentpm conv search "Alpaca live broker halt"
Agent
I found four related conversations and the latest outcome. Here's the thread.
Here's the thread
What happened
- The broker-connectivity halt was implemented and pushed.
- Trading now stops when broker state is unavailable.
- The previous agent verified host-local API health.
What remains
- Check Fresh/Alpaca authentication.
- Review PM2 logs and broker state before enabling orders.
Sources
4 conversations / 1 commit / 2 verification checks
I'm ready to continue from this point.
Alpaca halt
commit 6bbac50
host-local checks
#searchable history
#links back to sessions
#agent continuity
// captured as you work
The full work history, captured as it happens.
AgentPM captures conversations, commands, code changes, decisions, tests, and outcomes in one searchable history.
Humans can review the complete record. Future agents can retrieve the relevant conversations, decisions, and outcomes they needed.
Read the setup guideagentpm — shared work historyCollecting
Agent
Works
Prompts, edits, runs tools, and ships changes.
AgentPM
Keeps the work history
Conversations
Decisions
Code and tool activity
Tests and outcomes
Open work
Sources and outcomes
Next agent
Picks up the thread
Retrieves the relevant state before planning.
People
Search and review
Retrieved when needed
The complete history stays searchable. Relevant context is retrieved when needed.
// pick up the thread
The next agent starts where the last one stopped.
AgentPM brings forward the decisions, code changes, tests, risks, and open work from earlier sessions—so the next agent can continue from the known state instead of rebuilding context.
Continue from the latest verified outcome.
Carry decisions and unfinished work across agents, models, and tools.
Avoid repeating failed approaches or reopening settled questions.
Every captured session remains searchable. AgentPM retrieves what matters for the task at hand.
agentpm.dev / handoff contextready
User
1:32 AM
Check AgentPM for the Codex conversation
where I left off on the EMA Breakout feature.
What did we decide, what was still open,
and what should we not repeat?
Agent
1:35 AM
I'll recover the prior AgentPM thread
before planning.
Looking across the previous Codex and Claude
sessions for:
decisions · open work · warnings · failed paths · source turns
Context from prior threads
Recovered conversation state, linked back to the original sessions.
Known state
EMA Breakout was intended to stay a 15-minute strategy.
Last decision
Keep the preset behavior aligned across backend defaults, UI copy, and tests.
Watch out
A later session found the preset had drifted back to 5m in both backend defaults and UI.
Do not repeat
Do not redo the stale-preset investigation. That path was already completed.
User instruction
Do not redesign the strategy. Only fix the preset drift and verify the regression.
Next move
Check backend default, UI label, and regression test together before editing anything else.
// for humans
Read the insight. Back it up with the scrollback.
AgentPM distills every conversation into a one-line summary and typed keyframes you can filter: decisions, outcomes, risks, artifacts, config, state changes. Each one links to the exact turn that produced it.
See what moved forward across projects and agents.
Surface repeated friction, useful advice, and opportunities to improve.
Understand conversations, turns, tokens, and overall utilization at a glance.
Insights
Read typed keyframes, then jump back to the exact source turn.
// start with real work
Try AgentPM on the work you’re already doing.
Use your normal coding agents and workflow. AgentPM captures the sessions so people can understand what happened and future agents can pick up the thread.
No new workflow. Use the coding agents you already use.
Team
Share the thread across the team.
- Search across agents, developers, repos, and projects
- Give reviewers shared timelines, notebooks, and PR context
- Turn recurring friction and strong workflows into team learning
Enterprise
Control how AgentPM scales.
- SSO/SAML, RBAC, audit logs, and retention controls
- Data residency and private or VPC deployment options
- Govern sensitive work and advanced review workflows