Credit: Coworker.ai
Coworker.ai is launching OM2, a unified organizational memory layer that ingests data from 50+ enterprise platforms and distills it into precomputed, permission-aware facts. Instead of rebuilding context from scratch for every AI query, OM2 reuses structured intelligence across sessions, claiming a 9x reduction in token spend and 64% faster response times. Combined with optimized model routing, total cost savings can reach 51x.
Every time a typical enterprise AI agent spins up to answer an internal question, it wakes up with severe amnesia.
To make the AI useful, companies are forced into a costly, repetitive cycle: scraping thousands of documents, Slack threads, and CRM entries, and force-feeding them into the prompt window for context.
It is a brute-force approach to data retrieval that is estimated to consume over 50% of enterprise AI token budgets.
To solve this, Coworker AI is today launching OM2, a unified organizational memory layer designed to massively streamline how AI accesses corporate data.
Instead of rebuilding company context from scratch for every single session, OM2 continuously ingests data across more than 50 enterprise platforms and distills it into a living, permission-aware neural graph.
Traditional enterprise AI relies on document-level retrieval, pulling massive text chunks every time an employee asks a question. Coworker.ai’s OM2 takes a more surgical approach. As new information flows across the company, whether it’s a closed-won deal in Salesforce, a product pivot in Slack, or a strategic decision in a meeting transcript, OM2 extracts it into discrete, precomputed facts.
These facts map out exactly who was involved, what was decided, and which accounts are impacted. Because this structured intelligence is computed once and reused across queries, the AI doesn’t have to re-derive the context on every prompt.
According to the company, this structural shift alone drives a 9x reduction in token spend and speeds up response times by 64%, while delivering answers that users prefer on quality 84.5% of the time.
Coworker is also attacking the ‘one-size-fits-all’ model usage that inflates enterprise bills. By layering in the company’s Optimized Routing engine, OM2 dynamically triages tasks. It routes simple data retrieval to cheaper, faster models, while reserving heavy-duty reasoning for frontier models. Combined with the precomputed context layer, Coworker claims cost savings can reach a staggering 51x.
Crucially, this architecture grants companies true data sovereignty. Rather than locking their institutional knowledge inside a single vendor’s closed ecosystem, OM2 acts as an agnostic intelligence layer. Through its Model Context Protocol (MCP) or native apps, it plugs directly into the tools teams are already using: Claude, ChatGPT, Gemini, Perplexity, or custom-built internal agents.
“Token costs are exploding, and most AI still doesn’t understand how to be effective inside a real company,” said Alex Calder, Co-Founder and CEO of Coworker.ai. “OM2 is continuously learning and understanding your business, while still enforcing strict permissions at every node. That’s how we get to 9x cheaper before routing even enters the picture. And unlike a lot of the AI industry, we’re not incentivized to keep you burning more tokens. We’re incentivized to help you use fewer.”
In the enterprise space, intelligence is only as valuable as its security constraints; a junior engineer querying an agent shouldn’t stumble into the CEO’s M&A discussions.
OM2 addresses this by baking access policies directly into every fact and connection within the graph. Running on isolated, single-tenant infrastructure that meets SOC 2, GDPR, and CASA Tier 2 standards, the platform ensures that users and agents only ever extract the nodes they are explicitly authorized to see.
Early adopters are already utilizing the platform to bridge data silos. At RapidSOS, the Revenue Operations team previously spent hours manually cross-referencing Salesforce records with scattered communication logs to verify data. “Now the answer is already there, current and correct, the moment we ask,” noted Anna Waring, RapidSOS’s Director of Revenue Operations and Systems.
Backed by notable Silicon Valley investors including Ramtin Naimi and former Google SVP Jeff Huber, Coworker.ai is betting on a pragmatic shift in the AI landscape. While foundational model builders race to build ever-larger context windows, Coworker is proving that the real enterprise unlock isn’t necessarily a bigger brain, it’s a better memory.
Get the most important tech news in your inbox each week.
Contributed article. Not produced by the TNW newsroom and does not reflect the editorial stance of TNW.
