Evolution map & risk signals
Timeline, branch view, semantic diffs, and rule-based findings on each version—so you see drift before you need a rescue restore.
Checkpoint vault · Evolution observability
See how your agents evolve. Roll back when they drift.
Back up Persona and Work agents as versioned JSON from the runtime you already use. Compare checkpoints, spot risky edits, pin a stable baseline—and optionally share a redacted public page or template link.
Free: 1 agent, 3 versions, validation + evolution diff on every save, recovery export.
What you get
A vault for agent-generated checkpoints—not chat logs.
Rollback without erasing history
Restore an earlier checkpoint as a new version. Prior versions stay intact for audit and side-by-side compare.
Agent-generated checkpoints
The agent you already use writes structured JSON from its own context—Hermes, OpenClaw, companion sessions, or local tools.
Recovery you can export
Deterministic recovery prompts and Work recovery kits. Optional one-turn preview to sanity-check another model.
Persona & Work in one vault
Companions get identity and boundaries; work agents get skills, integrations, workflow rules, and procedural checklists.
Public showcase (optional)
Publish a redacted evolution timeline or single-checkpoint template—link-only or discoverable. Visitors can clone into their own private vault when you allow it.
How it works
Back up an existing agent in minutes—not from a blank form.
- 1
Back up an existing agent
Register the agent you already use. Choose Persona or Work—Agent Capsule does not create agents from scratch.
- 2
Run prompt in that agent
Open the agent you are backing up (Hermes, OpenClaw, a companion session, or local runtime) and run our checkpoint prompt there. Its LLM writes JSON from that agent's context—then paste it here.
- 3
Upload and review
Validate structure, fix gaps, and save an immutable version. Review fields in the editor when needed.
- 4
Restore elsewhere
Export recovery prompts or run a one-turn preview. Optional API and CLI for automated checkpoints on paid plans.
Two agent types
Persona continuity and work-agent recovery in one vault.
Persona Agent
Companions, roleplay characters, personal agents—identity, boundaries, memory anchors, and relationship continuity.
Work Agent
OpenClaw, Hermes, ops assistants—skills, tools, workflow rules, and a recovery checklist when runtimes reset.
How you checkpoint
Generate inside the agent you back up. Skill and CLI when you are ready.
- Available
In the agent you back up
Run the checkpoint prompt inside Hermes, OpenClaw, or the companion session itself. The runtime calls its LLM with that agent's files and memory—not Agent Capsule.
- Beta
Skill or CLI
OpenClaw/Hermes skills and the agentcapsule CLI push schema-valid JSON from the runtime side—without opening a separate chat.
- Available
Paste import (fallback)
Already have JSON or a character export? Upload here. Avoid using a fresh generic LLM thread as your only source—that rebuilds from snippets, not live agent state.
Built for
People who care about continuity—not another database of chats.
AI companions & roleplay
Keep one character’s persona safe when platforms, devices, or models change.
Creators with multiple agents
Protect a roster of agents with clear per-plan limits—not a folder of loose JSON.
Local & work agents
Back up OpenClaw, Hermes, and workflow assistants with Work checkpoints and optional API tokens.
Shipped today
Evolution observability—not just a folder of JSON files.
Every saved checkpoint gets a structured diff, risk level, and recovery-readiness score. Mark a version stable, compare two checkpoints, or generate a short LLM summary of what changed. When you are ready, publish a public-safe slice—private memory, recovery secrets, and scan output stay out.
How we think about the product →v01
Imported checkpoint
Validated JSON from the source runtime. Analysis: initial import.
v02
Review pass
Diff flags softer boundaries. Readiness moves from weak → usable.
v03
Stable baseline
Pinned as known-good. Later versions show drift vs this baseline.
Available today
Everything you need to try checkpoint backup.
- Checkpoint import (paste, upload, or API on paid plans)
- Persona and Work capsule types
- Evolution timeline, compare, and per-version analysis
- Stable baseline + drift indicators on the agent list
- Version history with pin and plan-based retention
- Recovery prompt export (text, Markdown, JSON)
- Monthly recovery-preview and evolution-summary quotas
Included on every plan
Validation is not a paid upsell.
- Schema validation on every import
- Structured diff + rule findings on each new version
- Missing-field and completeness hints
- Secret-risk warnings (no paid “validation unlock”)
- Recovery prompt export
Honest limits
Clear boundaries build trust.
- Checkpoints are structured summaries from the source runtime's context—not a byte-perfect server dump
- Not a generic vector memory or chat-log archive
- No automatic summarization from raw chat transcripts
- Recovery quality depends on what the runtime gave the LLM—you preview, not guaranteed clones
Paid plans add more protected agents and retained versions—not unlock validation. Compare plans.
FAQ
Common questions
- Does Agent Capsule store my full chat history?
- No. The vault stores structured checkpoints you or your agent produce—identity, skills, workflow, and recovery instructions—not entire conversation logs by default.
- Do I create a new agent inside Agent Capsule?
- You back up an agent you already use. You name it, import a checkpoint JSON, and keep versioned snapshots. Agent Capsule is not an agent builder.
- How is this different from a memory bank?
- Memory banks optimize retrieval from many messages. Agent Capsule optimizes continuity: immutable checkpoint versions, rollback, and standardized recovery exports.
- Who generates the checkpoint JSON?
- The agent runtime you are backing up—using its configured model and whatever context it loads (persona files, SOUL.md, skills, session memory). Agent Capsule validates and stores versions; it does not read Hermes or OpenClaw servers directly. Pasting our prompt into a separate chat with no agent context is only a fallback reformat, not a live self-backup.
- Is validation a paid feature?
- No. Completeness hints and secret-risk warnings are included on every plan. Paid tiers add more agents, retained versions, API checkpoints, and previews.
- Is there a free plan?
- Yes. Free includes one agent, three retained versions per agent, recovery export, evolution diffs on every save, and small monthly quotas for recovery preview and evolution summaries. See Pricing for Pro ($9/mo) and Pro Plus.
- Can I share an agent publicly?
- Yes—optionally. Publish a redacted checkpoint template or an evolution timeline page with only the versions you select. Private fields never leave your vault; you can unpublish anytime without deleting clones others already saved privately.
- Is public sharing a marketplace?
- No. There is no ranking, resale, or browse-all directory in this release—just link-based pages you control, with optional unlisted (noindex) mode and a report button for abuse.
Keep the checkpoint. Lose the scramble.
Back up an agent you already use. Watch versions evolve, roll back when something drifts, and share a redacted template only if you want to.