← all repositories
RyjoxTechnologies/Octopoda-OS

A persistent memory layer so agents survive crashes and loops

Octopoda gives Python AI agents persistent memory that survives crashes, detects costly loops in seconds, and exposes a live dashboard—running locally with zero infrastructure.

501 stars Python LLMOps · EvalAgents
Octopoda-OS
Collecting fresh signals — velocity needs a few days of history.
collecting data…
star history

What it does Octopoda is a Python middleware layer that slips under agents built with LangChain, CrewAI, AutoGen, OpenAI’s SDK, or plain Python. It persists memory across restarts and crashes, versions every write automatically, and runs a five-signal loop detector that flags retry, oscillation, and ping-pong patterns. A built-in local dashboard—or a cloud-hosted equivalent—shows fleet health, memory history, and audit trails in real time.

The interesting bit The audit trail can be hash-chained per agent, giving you tamper-evident logs of every decision and its memory snapshot without an external ledger. Multiple agents can also share atomic memory spaces, with every write tagged to its author and conflicts surfaced for review.

Key highlights

  • Survives process restarts, crashes, and deploys using local SQLite by default, or PostgreSQL/pgvector when synced to cloud.
  • Catches five loop signatures—retry, oscillation, ping-pong, reflection, and recall-write—and surfaces the exact calls responsible.
  • Audit trail captures a memory snapshot at every decision and can hash-chain events for verifiable provenance.
  • Cross-agent shared memory tracks authorship and timestamps; concurrent writes to the same key surface as conflicts.
  • Same Python API for local and cloud; moving from offline to synced needs only an environment variable change.

Caveats

  • Loop detection runs automatically, but stopping an agent via auto-pause or spend caps is opt-in policy under the v2 circuit-breaker.
  • Shared memory resolves concurrent writes to the same key with last-write-wins by default, so unreviewed conflicts can silently overwrite prior values.
  • Offline semantic search is excluded from the base package; it ships as an optional ai module weighing roughly 33 MB.

Verdict Reach for Octopoda if you are shipping Python agents to production and need to debug loops, audit decisions, or persist state without managing Redis. Skip it if you are only calling an LLM in a stateless script and do not care what happens after the process exits.

Frequently asked

What is RyjoxTechnologies/Octopoda-OS?
Octopoda gives Python AI agents persistent memory that survives crashes, detects costly loops in seconds, and exposes a live dashboard—running locally with zero infrastructure.
Is Octopoda-OS open source?
Yes — RyjoxTechnologies/Octopoda-OS is an open-source project tracked on heatdrop.
What language is Octopoda-OS written in?
RyjoxTechnologies/Octopoda-OS is primarily written in Python.
How popular is Octopoda-OS?
RyjoxTechnologies/Octopoda-OS has 501 stars on GitHub.
Where can I find Octopoda-OS?
RyjoxTechnologies/Octopoda-OS is on GitHub at https://github.com/RyjoxTechnologies/Octopoda-OS.

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.