frege

the memory layer for your AI agents.

Your agents are either flying blind or seeing what they shouldn't. Frege gives every agent exactly the context it's allowed to have — versioned, audited, revocable. One source of truth for Codex, Claude Code, and everything next.

agent@frege:~$ claim your pilot seat

agent@frege:~$ cat problem.txt

# teams adopt agents faster than they standardize the memory those agents need. every team keeps rebuilding the same thing.

"every team is building their own context folder for their AI."
"we don't know what our agents are reading."
"we don't want Codex to see HR documents."

agent@frege:~$ frege --help

frege turns institutional knowledge into permission-aware infrastructure.

memory: one source of institutional knowledge for every agent stack
versions: versioned markdown with a real audit trail
access: role-based control over what each agent can read or write
mcp: MCP-native and model agnostic — the same layer for every stack
usage: audit logs, usage, and estimated cost by user, key, and org

agent@frege:~$ frege whois

founders: every agent on the same trusted knowledge, without exposing everything
CTOs: one controlled place agents read from, not a folder per harness
platform leads: a memory layer that works across Codex, Claude Code, and internal agents
security leads: access control and a clean audit trail over what an agent did

agent@frege:~$ frege pipeline --explain

[1] store: bring your markdown. it becomes versioned company knowledge.
[2] govern: define orgs, roles, and API keys. RBAC decides who reads or writes.
[3] connect: agents reach approved context through MCP tools, across any stack.
[4] audit: every access is logged. review usage and cost by user, key, and org.

agent@frege:~$ frege read --as claude-code@platform

read /eng/runbooks/deploy.md: [granted]
read /eng/architecture/v3.md: [granted]
read /hr/compensation.md: [denied: role]
edit /eng/runbooks/deploy.md: [proposed: v14]

agent@frege:~$ cat why-now.txt

agent adoption is outpacing the memory layer.

companies are adopting agents faster than they are standardizing the knowledge those agents need. every team rebuilds the same internal infrastructure — a markdown folder for one harness, a different setup for the next, ad hoc rules about what an agent can read, and no shared audit trail for who or what touched institutional context.

the issue is not that markdown files are hard. it is that institutional context becomes unsafe, fragmented, and expensive to maintain when every team invents its own agent-readable knowledge layer. the job now is one controlled place to read, write, version, and audit institutional knowledge — before the next agent gets there first.

agent@frege:~$ frege pilot --join

We are taking a limited founding pilot. Founding pilots help shape the roadmap, get hands-on onboarding from the team, and lock in early pricing before general availability. Bring your agents; we'll bring the memory layer.

agent@frege:~$ claim your founding pilot seat →

seats are limited · no credit card · 2-minute form

agent@frege:~$ cat privacy-policy.txt

# frege privacy policy · last updated 2026-06-07. plain language, no dark patterns.

what we collect: only what you type into the early-access form: name, work email, company, role, company size, expected users, current agent tools, monthly AI spend, what you'd expect to pay, decision timeline, your main pain point, and any optional comments. nothing else.
what we never collect: we do not ask for, store, or transmit your company documents, source code, customer data, or API keys through this site. there is no tracking pixel, no third-party analytics, and no advertising cookie on this page.
why we collect it: to evaluate early demand, gauge fit for a pilot, and reach out about early access. that is the only purpose.
how it is stored: submissions are written to a private Postgres database with access limited to the Frege founding team. data is encrypted in transit (TLS) and at rest.
sharing: we do not sell your data and do not share it with advertisers. limited processors (database host, email) act on our instructions under contract.
retention: we keep early-access submissions until the validation phase ends, then delete or anonymize records we no longer need.
your rights: email hello@frege.dev any time to access, correct, or delete your data, or to opt out of contact. we will action it.
contact: hello@frege.dev

agent@frege:~$ frege compliance --status

# security and compliance posture · validation stage.

soc 2: Frege is pre-certification. SOC 2 Type II is on our roadmap and we are building the product to SOC 2 control objectives — access control, encryption, audit logging, and change management — from day one. We will pursue a formal Type II audit before general availability.
today: data in transit is encrypted with TLS; data at rest is encrypted; access to early-access submissions is restricted to the founding team and logged.
by GA: SOC 2 Type II report, role-based access control, full tenant isolation, and per-org audit trails. permission-aware access and audit logging are core product features, not afterthoughts.
questions: security or compliance diligence: hello@frege.dev

This statement describes our current posture and forward-looking plans. It is not a representation that Frege currently holds a SOC 2 certification.