Public product roadmap

Governed memory first. Controlled action follows.

Your agents keep their own models and execution environment. Frege is shipping the governed organizational-memory layer they share, then adding policy-controlled integrations without weakening provenance or customer choice.

Available
Supported in the current product.
Beta
Usable now; interfaces may change.
Planned
Direction, not a delivery-date commitment.
A stone facade centered on a sundial and arched doorway, rendered with numeric glyphs.
Status over timeNumeric-character study / source 03

Use today

Available now

The governed memory and context foundation available to Frege customers today. Beta capabilities are usable, but their interfaces may still change.

Available

Versioned organizational memory

Store organizational knowledge as governed pages and sources with revision history instead of scattering it across agent-local files.

Available

Scoped, cited context

Build task-specific context from sources the caller is allowed to receive, with citations back to the returned material.

Available

Reviewable memory updates

Let agents propose durable knowledge changes while keeping canonical memory under human review.

Available

Session and event ledger

Keep durable task context, context builds, tool activity, and notes connected to the work that produced them.

Beta

Shared live Codex sessions

Authorized teammates can watch, redirect, stop, resolve exact approval requests, and hand off a live Codex run while execution stays on its originating machine.

Available

MCP and CLI access

Connect Codex, Claude Code, and other MCP clients through the Frege CLI and a scoped API key.

Available

Customer-run agents

Connect Codex, Claude Code, or an internal agent—or download the Frege Agent profile for Hermes—while model access, tools, and compute stay in the customer's environment.

Policy first

Building next

The identity, policy, and provenance controls Frege needs before agents can safely receive broader access and take more consequential actions.

Planned

Service principals

Represent humans and services as explicit identities with scoped, delegable credentials in the V2 technical preview. Direct identities for customer-run agents remain planned.

Planned

Versioned authorization policies

Evaluate principal, action, and resource together under default-deny policies that can be reviewed and versioned.

Planned

Authorization receipts

Return an attributable record of why access was allowed or denied without revealing restricted source content.

Planned

Governed connector pilots

Pilot repository-scoped GitHub sync only after the identity, recovery, deletion, and provenance contract is ready. Google Drive follows later.

Planned

Unified provenance ledger

Connect reads, denials, context builds, sessions, tool activity, proposals, and approvals in one attributable event history.

Deliberate expansion

Later

Broader workflow and optional execution capabilities built only after the memory product and its policy boundaries are proven.

Planned

Durable tasks and workflows

Track assigned principals, steps, inputs, outputs, and idempotent state transitions without overloading agent sessions.

Planned

Durable workflow approval gates

Extend the live Codex approval path into durable, multi-step workflows that pause, collect an authorized decision, and resume from recorded state.

Planned

Optional hosted execution

Offer bounded execution only when customers need it, under the same capability, approval, step, and spend controls as every other agent.

Planned

Portable organizational record

Export and import organizational knowledge with source, revision, and provenance context so customer memory is not trapped in one runtime.

Planned

Stable control-plane contracts

Add versioned API contracts for principals, policies, receipts, connectors, tasks, and approvals while preserving current v1 clients.