CClawMetryDocs

Runtime reference

Replit Agent#

Replit Agent's loop runs on Replit's infrastructure, but it serializes its working state into the Repl's own filesystem — which you own, can SSH into, can clone, and where a ClawMetry daemon can run. That on-disk state is what the adapter reads; there is no vendor API call anywhere in it.

Runtime idreplit
CategoryCloud app builder
VendorReplit
TierPaid tier (Starter and up)
Upstreamreplit.com

Where the data comes from#

ClawMetry reads these locations, in order, and never writes to any of them.

  • .local/state/replit/agent/transcript/<session-uuid>/transcript.jsonl — one journal per session
  • .local/state/replit/agent/repl_state.json — the Repl description
  • .local/state/replit/agent/.latest.json — the current branch
  • .local/state/replit/agent/progress_tracker.md — the agent's own checklist
  • replit.md — agent instructions and memory

Format. Each transcript.jsonl line is one flushed batch {"memory_id", "messages"}. Batches are deltas and never overlap, so concatenating messages in file order yields the full conversation. Messages are Anthropic-shaped.

Environment variables#

VariableEffect
CLAWMETRY_REPLIT_ROOTSworkspace roots, os.pathsep or comma separated; when set it is the ONLY source

See Environment variables for the variables that apply to every runtime.

Tokens and cost#

Tokens. None. No tokens, no model names and no dollars exist anywhere in this store.

Cost. Unavailable. Replit bills effort-based checkpoints rather than tokens, so cost_usd is None with cost_status="unavailable" and COST is not claimed.

Why this section exists

Runtimes differ enormously in what they persist. Rather than showing a plausible number everywhere, ClawMetry reports what the store actually contains and marks the rest unknown. A cost of $0.00 and a cost of unknown mean different things, and the dashboard keeps them different.

Sub-agents and delegation#

Not exposed. The journal records one conversation per session with no delegation structure.

Things worth knowing#

  • The .agent_state_*.bin files are an opaque serialization and are deliberately never parsed. A format guess that broke on the next Replit release would take the whole adapter down with it.
  • No timestamps exist inside the journal. Session times come from the transcript file's mtime and say so via extra.timeBasis="file_mtime". Inside a live Repl those are real write times; in a git clone they are checkout times — declared either way, never dressed up.
  • There is no pid to signal: the agent loop runs on Replit's servers, not in the workspace container. Process control carries the matching refusal string rather than a button that quietly does nothing.
  • Inside a live Repl the roots are found with zero configuration from $REPL_HOME / $REPL_ID. Outside one, point CLAWMETRY_REPLIT_ROOTS at the clone.
  • User text arrives wrapped in <user_message> with injected <attached_contents> blocks. Titles come from the typed span only — using the whole wrapper titles every session with injected context.

Verify it is being read#

bash
# Is this runtime detected, and would this install unlock it?
clawmetry runtimes --json | jq '.runtimes[] | select(.id=="replit")'

# Why is it locked, if it is?
clawmetry runtimes --why replit

# Any sessions ingested yet?
curl -s 'http://localhost:8900/api/local/sessions?limit=5' \
  | jq '.rows[] | {session_id, agent_id, events, cost_usd}'

If detection succeeds but no sessions appear, work through Troubleshooting — the usual causes are a relocated data directory (set the override above), a session limit that is hiding older runs, or a runtime that simply has not written anything yet.

Edit this page View as Markdown Report an issue
Cookie preferences