The team aligned on shipping the agent retrieval API by Q3 and re-pricing the Builder tier. Marketing will lead an external launch the week of Sept 23. Engineering owns the latency target of <40ms p95.
- Launch window: week of Sept 23
- Builder tier moves to $19/mo
- Latency target: <40ms p95
- Owner: @maya (product), @jonas (eng)
- 00:04 — Kickoff & objectives
- 00:18 — Pricing v2 proposal
- 00:39 — Latency budget review
- 00:51 — Action items locked
AI has an amnesia problem
AI forgets everything after each conversation
Context disappears between sessions
Projects lose continuity and momentum
Teams repeat themselves endlessly
Agents restart from zero every time
Persistent memory for every AI
RecallOS creates a unified memory layer that stores, indexes, and retrieves information across every conversation, project, and workflow.
Store every conversation permanently
Maintain full project context
Build cumulative knowledge bases
Share memory across team members
Give agents persistent intelligence
A single pipeline for memory and intelligence.
From signal to response — every step instrumented, observable, and swappable.
Everything an agent needs to actually remember.
Long-lived storage that survives sessions, processes, and deploys.
Meaning-aware retrieval powered by hybrid vector + keyword indexes.
Adaptive context windows tuned per model and per task.
Scope memory by project, customer, or environment.
Drop-in tools your agents can call to recall, store, or forget.
Pick up exactly where you — or your agent — left off.
Importance and recency scoring keeps the right facts on top.
Run RecallOS locally, on prem, or as a managed service.
Works with any LLM — closed, open, or self-hosted.
Plug RecallOS into the model you already use.
One memory layer, every model. Switch providers without losing a single thought.
Anthropic's reasoning workhorse.
OpenAI multimodal series.
Google's long-context family.
High-throughput open weights.
Alibaba's multilingual stack.
Meta's open-weight backbone.
Compact, fast, Euro-built.
Inference for any open model.
Build your memory layer.
One command to install. One command to initialize. Your knowledge becomes persistent forever.
Built for production. Measured in production.
Independent runs across 1.2B tokens of real workloads.
Multi-region replication, point-in-time recovery, and per-tenant memory isolation by default.
Trusted by builders shipping real agents.
"RecallOS turned our copilot from amnesiac to actually useful. The latency budget is real and the recall is uncanny."
"We replaced 600 lines of brittle context-stitching with a single recall() call. Our agents stopped lying about prior runs."
"It's the first memory product that feels designed for production, not a demo. Importance scoring is a quiet superpower."
"Cross-session recall is the feature I didn't know I was missing. My customer-support agent finally has a memory."
"Open source, model agnostic, dead simple SDK. RecallOS is what every team should reach for first."
"We benchmarked four memory layers. RecallOS won on accuracy and latency at the same time."
Simple, predictable, scaled to your memory.
For solo builders exploring agent memory.
- 100k memory entries
- Semantic + keyword recall
- Single project
- Local + cloud SDK
- Community support
For teams shipping production agents.
- 10M memory entries
- Importance + recency ranking
- Unlimited projects
- Cross-session recall
- Priority email support
- Audit log
For platforms with memory at scale.
- Unlimited memory
- Dedicated regions
- SSO + SCIM
- On-prem / VPC option
- 99.99% SLA
- Solutions engineer