fix: upsert traces to handle duplicate IDs from intermediate flushes

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-Claude)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

launch/twitter.md

@@ -0,0 +1,67 @@
+# AgentLens Launch -- Twitter/X Thread
+
+---
+
+**Tweet 1 (Hook)**
+
+Current agent observability tools tell you WHAT API calls your agent made.
+
+They don't tell you WHY it picked tool A over tool B, or why it retried instead of escalating.
+
+That's the gap I kept hitting. So I built something to fix it.
+
+---
+
+**Tweet 2 (What it does)**
+
+AgentLens traces agent decisions, not just LLM calls.
+
+It captures tool selection, routing, planning, retries, and escalation -- with the reasoning, alternatives considered, and confidence at each step.
+
+Open source. MIT licensed. Built solo in 2 weeks.
+
+#AI #OpenSource #Agents
+
+---
+
+**Tweet 3 (Code)**
+
+Four lines to get started:
+
+```
+pip install vectry-agentlens
+
+import agentlens
+agentlens.init(api_key="key", endpoint="http://localhost:4200")
+wrap_openai(openai.OpenAI())
+```
+
+Auto-instruments your OpenAI client. Then trace decisions as they happen.
+
+---
+
+**Tweet 4 (Features)**
+
+What you get:
+
+- Live Next.js dashboard with decision flows
+- OpenAI auto-instrumentation via wrap_openai()
+- 7 decision types: routing, planning, tool selection, retry, escalation, memory retrieval, custom
+- Self-host with Docker Compose
+- Python SDK on PyPI
+
+#DevTools #LLM
+
+---
+
+**Tweet 5 (CTA)**
+
+AgentLens is v0.1.0 -- early but functional. Rough edges exist.
+
+Try the live demo: https://agentlens.vectry.tech
+Repo: https://gitea.repi.fun/repi/agentlens
+Install: pip install vectry-agentlens
+
+Feedback welcome, especially on the decision model.
+
+#OpenSource #AI #Agents