Trust Runtime · Engineering View
Starter path · Operational trust first
Install, run, inspect—then replay
“Start free” is for builders who want receipts: trust scores, Evidence (GhostLog) lineage, timelines, and guided missions—not a conceptual tour first. Finish with Recon vs chaos when you want the why behind orchestration, or read the runtime continuity manifesto for the replay-native posture behind it.
Run your first replay in under 5 minutes
Generate local trust receipts, open seeded replay, then spot drift, interventions, and autonomy decisions—labeled honestly as starter/demo workspaces until ingestion is wired.
Run first replayActivation in ~5 minutes
Keep your runtime. Add receipts.
Add the SDK, then run the local demo path before wiring production ingestion.
Jump to install →Use the deterministic playbook to see trace, replay, drift, and evidence vocabulary.
Open first replay →These surfaces may be empty until ingestion is wired; seeded replay stays available.
What you'll see in a wired workspace (Starter inspect-only posture):
- ▸Replay reconstructed from the receipts Recon can see.
- ▸Starter posture stays inspect-only until you wire authority.
- ▸Evidence continuity shows where lineage, drift, and checkpoints held.
Install ReconAI in under 60 seconds.
Detect silent AI failures before they become production incidents.
Step 1 · Install the SDK
TypeScript / Node surface. LangChain adapter (Python): reconai-langchain on GitHub (PyPI when published). Explore version and peer deps on npm.
Step 2 · See Recon detect drift
Run a local demo to watch trust scores, drift signals, and GhostLog receipts appear in your terminal.
What you'll see in 60 seconds
- Healthy baseline
- Silent drift
- Tool misuse
- GhostLog replay
- Trust score
What Recon surfaces that logs miss
- Model output looks correct
- Weak evidence chain
- Policy misalignment
- Confidence exceeds evidence
- Reflex Score 45
- ▸Fastest scripted path on the web: First replay playbook
- ▸In this monorepo, from
packages/sdk:pnpm run example:quickstart·pnpm run example:langchain
Step 3 · Inspect trust in the product
TimelineSee trust evolve over time
Open timeline →
GhostLogReplay exactly what happened
Open ghostlog →
ReplayWalk the incident step-by-step
Open replay →
AutonomyKnow when agents should pause
Open autonomy →
Step 4 · Why Recon catches failures other tools miss (optional)
Same workflow with and without the trust layer—use this after you have a mental model from installs and replay. Better as mid-funnel conviction than first click.
See Recon vs chaos (narrative)Ready to monitor a real agent?
Create a free workspace to persist GhostLogs and trust events, or keep exploring with the SDK.