Actx0

What is Actx0?

Actx0 is managed memory infrastructure for AI agents and apps. It sits as two calls around your model — you still run the model yourself — storing session turns, extracting them into short searchable facts in the background, and returning the relevant hits to inject into the prompt before the next reply. The point is to stop replaying the full transcript on every turn without having to operate a vector store: the site frames it as "a few facts, not the whole chat." It runs only as managed cloud, with workspaces, roles, access keys, audit events and usage tracking for teams.

Key features

• Three-step loop documented as Add / Learn / Retrieve: you POST user messages into an agent session, Actx0 stores and indexes the turn, background extraction turns useful content into short facts (greetings are skipped) and consolidates them within the session, and you search before generating a reply.
• Managed semantic search across both session memories and workspace knowledge (your documents), so a question about your docs and a question about the conversation hit the same retrieval call.
• Deduplicates and consolidates memories within a session rather than accumulating raw turns.
• Python, Node.js, Go and REST clients — the site's stated position is that there is no vector store for you to operate. The Python client installs as `pip install pctx0` and exposes agent.create, session.create, message.create and memory.search.
• Workspace controls aimed at teams: workspaces, roles, access keys, audit events and usage tracking, with isolated memory namespaces and per-team billing on shared infrastructure.
• Fully managed cloud — the vendor's framing is that there are no servers for you to provision, patch or scale.
• Self-serve monthly billing with a free tier: Hobby $0/mo, Starter $19/mo, Growth $69/mo, Pro $219/mo, Enterprise custom.
• Headline performance figures published on the site — under 10 ms P99 retrieval latency and 90% token reduction — are vendor claims, not independently measured.

Who it's for

• Chat or agent products where the current workaround is stuffing the whole prior transcript back into every request, and the token bill grows with conversation length.
• Long-running assistants that need facts about a user to survive across sessions, agents and applications rather than dying with the session.
• Teams that would otherwise stand up and maintain their own vector store plus extraction pipeline just to get durable agent memory.
• Support, CRM and internal-tool assistants that must answer from both the conversation so far and a body of company documents — the site lists Customer Support, Sales & CRM, Healthcare, Education, DevTools and E-Commerce as its target use cases.
• Multi-tenant products that need per-team memory isolation, access keys and an audit trail over who read or wrote memory.

When not to use it

It is managed cloud only — the site advertises no self-hosted or in-VPC deployment — so if your conversation data cannot leave your own infrastructure for regulatory or contractual reasons, this is the wrong layer.

FAQ