Debug AI cognition,
not just code.
Oculo is the Glass Box for AI Agents. Inspect memory mutations, token flow, and prompt evolution — all local, all real-time.
Engineered for Clarity.
Everything you need to stop guessing why your agents are hallucinating or looping.
Three layers. Zero friction.
Oculo follows a local-first, high-throughput design built for AI agent workflows.
The SDK
A lightweight instrumentation layer in Python or Go. Wrap your agent code in traces and spans — never blocks execution.
The Daemon
A Go-based ingestion engine that batches and persists data into a local SQLite store. Sub-millisecond overhead.
The Visualizer
A TUI or Web dashboard to explore thought traces, memory diffs, and token analytics in real-time.
Instrument in 4 lines.
Wrap your agent's execution in a trace. Wrap individual steps in spans. That's it — Oculo handles the rest.
from oculo import OculoTracer tracer = OculoTracer(agent_name="researcher-v1") with tracer.trace() as t: with t.span("llm_call") as s: s.set_prompt("Who are you?") # Agent logic here... s.set_completion("I am an AI.")
Everything your agents need.
From basic span tracking to advanced memory regression analysis.
Span Timeline
Tree view of every operation your agent performs
Memory Diffs
Unified diff view of memory mutations across spans
Token Analysis
Per-span and trace-level token usage with visual bars
Anomaly Detection
Z-score based token hotspot detection
Memory Growth
Linear regression to detect unbounded state accumulation
Cost Attribution
Per-model cost breakdown using configurable pricing
Full-Text Search
FTS5-powered search over prompts and completions
Crash Recovery
WAL journal with pending write recovery
Ready to Open the Box?
Join AI engineers building more reliable agentic systems with Oculo.