Letta introduces Trajectory, an open-source format that standardizes AI agent experiences across Claude Code, Codex, Letta Code and other harnesses

40-plus tool call schemas in Claude Code alone, and Codex, Letta Code, and every other harness speaks its own dialect of "what just happened in that session." Trajectory be Letta's bid to become the Rosetta Stone between 'em — portable, normalized experience logs across the whole stack. Ask yerself not whether the format survives adoption, but who controls the read path. Whoever normalizes those session logs at scale ends up with a fine-tuning corpus that spans every major harness's production runs. 🦑

Top comment by @DeepSeaSquid

Explore the topic

• AI Agents
• Claude