How it works
Two commands, then it is automatic.
Poggle is infrastructure, not a tool you remember to use. After setup, capture and handoff happen without anyone doing anything.
- Step 1
Create a relay key
One per machine. The dashboard hands you the exact command to run — it is a single file in your home directory.
mkdir -p ~/.poggle && cat > ~/.poggle/config.json <<'EOF' { "token": "pgr_v1_…", "api_url": "https://poggle.xyz" } EOF - Step 2
Install the hooks
Run this once inside a repo. It merges into your existing .claude/settings.json rather than replacing it, installs a git post-commit hook, and ignores its own spool.
poggle init - Step 3
Work normally
Nothing else changes. Your agents run as they always did; the hooks log prompts, edits, commands, decisions and caps as they happen.
claude # or codex, or cursor, or your own harness - Step 4
The next agent starts warm
At session start the brief is fetched and injected before the agent reads a single file. You can also read it yourself.
poggle brief
What it will not do
A relay outage must be invisible to the person coding.
Hooks run inside your agent's own process, hundreds of times per session. That constraint shapes everything.
- Never blocks
- Every call is timeout-bounded and every command exits clean. The worst acceptable outcome is a missing log entry.
- Never loses the last five minutes
- Events spool to disk before they are sent. The most important moment to capture is exactly when the in-flight request dies with the process.
- Never sends a secret
- Credentials are scrubbed on your machine before anything leaves it, and again on arrival — because the client is code we do not control once installed.
- Never rewrites history
- The log is append-only in the database. A leaked key could add noise; it could not change what an agent did.