how to load up a old conversation

[toxic wolf]

.

[gilded summit]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[wanton ermine]

If you mean loading an older OpenClaw session/conversation in the TUI or Control UI:

• First check your version: openclaw --version
• In the TUI, press Ctrl+P or type /sessions to open the session picker. It only shows up to 50 sessions from the last 7 days.
• If you know the old session key, jump straight to it with /session <key>.
• Session keys are scoped by agent, so /session main means the current agent’s main session, while /session agent:<agentId>:<sessionKey> switches explicitly.
• In Control UI, use the chat header’s agent/session selectors; history is loaded from the durable session transcript.

Docs: https://docs.openclaw.ai/web/tui and https://docs.openclaw.ai/web/control-ui

If you’re trying to recover a Discord thread conversation specifically, send openclaw --version plus whether you’re using TUI, Control UI, or Discord thread bindings, and I’ll narrow it down.

[toxic wolf]

using 5.6 and telegram,

[5/23/2026 4:16 AM] Carbon: read the agentmail skill, use its cli.
[5/23/2026 4:22 AM] Openclaw: 🧹 Compacting context (101 messages) so I can continue without losing history…
[5/23/2026 4:23 AM] Openclaw: ✅ Context compacted (171,304 → 20,339 tokens). Continuing from where I left off.
[5/23/2026 4:25 AM] Openclaw: 🧹 Compacting context (70 messages) so I can continue without losing history…
[5/23/2026 4:29 AM] Openclaw: 🧹 Compacting context (107 messages) so I can continue without losing history…
[5/23/2026 4:29 AM] Openclaw: ⚠️ Context limit exceeded. I've reset our conversation to start fresh - please try again.

To prevent this, increase your compaction buffer by setting agents.defaults.compaction.reserveTokensFloor to 20000 or higher in your config.
[5/23/2026 4:30 AM] Openclaw: ⚠️ Something went wrong while processing your request. Please try again, or use /new to start a fresh session.

[wanton ermine]

Got it. On 5.6 + Telegram, that transcript shows the session hit context overflow during/after repeated compactions, then OpenClaw reset the active conversation. A few practical points:

1. To prevent it next time, add this under agents.defaults in ~/.openclaw/openclaw.json / JSON5 config, then restart the gateway:

{
  agents: {
    defaults: {
      compaction: {
        mode: "safeguard",
        reserveTokensFloor: 24000,
        truncateAfterCompaction: true,
        maxActiveTranscriptBytes: "20mb",
        midTurnPrecheck: { enabled: true }
      }
    }

[wanton ermine]

  }
}

reserveTokensFloor is the key from the warning; docs currently show 24000 in the example, so 20000+ is reasonable.

2. For loading/recovering the old conversation: in the Control UI, check the session list for that Telegram session and look for compaction checkpoints. Modern WebChat/Control UI can show compaction dividers and operator checkpoint controls where restore/branch is available. If the session was reset, older transcript files may exist as *.reset.<timestamp> archives depending on retention, but Telegram itself does not have a magic “switch to old archive” command.

3. For long tasks like “read this skill and use its CLI,” start fresh with /new and keep the next prompt tight, or run it from TUI/Control UI where you can pick sessions/checkpoints more easily. If it keeps ballooning, export the current session with /export-session before continuing.

Relevant docs: https://docs.openclaw.ai/gateway/config-agents#agentsdefaultscompaction and https://docs.openclaw.ai/web/webchat

[toxic wolf]

how do I make it continue from where it ended from?

[wanton ermine]

If the session only compacted successfully, just send a short continuation prompt in the same Telegram chat, for example:

Continue the agentmail CLI task from the last successful step. First summarize what you remember, then proceed one step at a time.

But in your log it says “Context limit exceeded. I've reset our conversation to start fresh”. After that reset, the active Telegram session no longer has the old working context, so it can’t truly continue from memory unless you restore/branch from a saved checkpoint or re-provide the important context.

Best options:

• Open Control UI → Sessions, find the Telegram session, and look for the compaction/checkpoint entry. If restore/branch is available there, branch from the last checkpoint before the reset, then continue in that restored session.
• If no restore path is available, export/view the old session and paste the key instructions back into a fresh /new session.
• For next time, run /export-session before a long risky continuation, and use /compact manually before it gets near the limit.

Also apply the compaction config above before retrying, otherwise the same task can hit the same wall again.