9D Labs provides Agent‑Core, a memory runtime that governs context selection, task state tracking, and policy enforcement for long‑running AI agents. The platform supplies APIs and a dashboard that let operators define token budgets, apply approval rules, and receive auditable receipts for every memory decision, enabling reproducible debugging and compliance. It monetizes through subscription access to the runtime, SDK, and console for enterprise teams deploying persistent agents.
Funding
Funding not disclosed
Founders
Product
Problem
Long‑running LLM agents frequently lose track of relevant context, exceed token budgets, and operate without transparent governance, leading to silent failures and costly post‑mortems. Teams must manually prune memory, stitch together task state, and retrofit audit logs, which scales poorly as agents become more autonomous.
Solution
Agent‑Core from 9D Labs delivers a managed memory runtime that sits between an agent and its knowledge store, automatically ranking and pruning memories to stay within token limits while honoring operator‑defined policies. The platform continuously tracks task state across retries, handoffs, and approvals, exposing a unified view in a web console that mirrors the runtime used by the agents. Before any action is executed, configurable policy checks—such as scope restrictions, risk‑based approvals, or kill‑switch overrides—are applied, ensuring compliance by design. Every memory inclusion, exclusion, and prioritization decision is recorded in an immutable receipt, enabling reproducible debugging and auditability. Access is provided via a RESTful SDK (MemoryClient) and language‑agnostic libraries, supporting safe, team, or autonomous trust modes for flexible deployment. The solution reduces manual prompt engineering, eliminates context‑drift failures, and shortens incident‑response cycles for production AI systems.
Target Audience
The primary customers are enterprise AI engineering and operations teams that deploy production‑grade, long‑running LLM agents requiring strict governance, auditability, and token‑budget management.
Features
- Automated context selection and token‑budget pruning based on relevance scoring and operator rules
- Persistent task‑state tracking that records active, blocked, completed, and failed steps across long‑horizon runs
- Pre‑execution policy engine with scope validation, risk‑level approvals, and emergency kill‑switch controls
- Immutable decision receipts detailing which memories were used, trimmed, or rejected for full audit trails
- SDK (MemoryClient) and HTTP‑standard library with optional extras for seamless integration into existing agent frameworks
- Real‑time dashboard showing memory packs, task timelines, policy outcomes, and system health in a shared operator view
- Workspace‑scoped memory artifacts and granular access controls for multi‑team environments
- Configurable trust modes (safe, team, autonomous) that embed capability checks into connection adapters