Roadmap
From gateway to enforcement.
Ordered by what unblocks real use first. Targets, not promises.
Done · v0.7 · Nov 2026
Automatic tightening
- Opt-in: stricter suggestions apply themselves, up to a number a day (Team, Enterprise)
- Each automatic change is watched, and undoes itself if it blocks more than you allow
- An undo touches only that change; edits people made since stay
Done · v0.6 · Nov 2026
Rules that improve
- Test any policy change on the last 30 days of traffic before saving it
- Every change kept as a version, with who made it; restore any one in a click
- The policy advisor (Team, Enterprise): suggested rules from people's decisions and paid-call results, each tested first, with shadow tests
Done · v0.5 · Nov 2026
Open source
- The gateway and console open source under the Apache 2.0 licence
- Single sign-on, audit export, metrics and the advisor in Blend0 Enterprise, for Team and Enterprise customers
Done · v0.4 · Nov 2026
Plans
- Free, Team and Enterprise plans, with licence keys checked on your machine
- Plan and seats in the console; a licence never affects agent traffic
Done · v0.3 · Nov 2026
Admin console
- An admin console: overview, approvals, activity, agents, policy, users and settings
- Users with roles: viewer, approver, admin
- Single sign-on with OpenID Connect, with groups mapped to roles (Enterprise)
- Policy and agents edited live, validated, with no restart
- Activity search, with export to CSV and JSON lines (Team)
- Every sign-in and admin change in the audit log; Prometheus metrics (Team)
Done · v0.2 · Nov 2026
Agent Gateway
- Host allow and deny rules, plus a check of the address each name resolves to
- Reads x402 payment requests and picks the cheapest acceptable option
- Per-request cap, daily budget, review threshold, allowed networks and currencies
- Per-agent keys and budgets, never looser than the global limits
- Trust scores from your settings and Blend0's own record of each service
- Single-use, expiring, signed approvals that survive restarts
- Approval notifications by webhook, and approving from the command line
- Append-only audit log; spend survives restarts
- Responses stream through untouched, binary included
- One command for Claude and other MCP clients, a drop-in fetch, and a proxy mode for HTTP(S)_PROXY
- A Docker image and a lockdown recipe that makes Blend0 the only way out
- Safe with no setup: cloud metadata blocked, an admin token made on first run
Next
Enforcement
| Item | Why |
|---|---|
| Wallets that check approvals | Enforce policy where the money moves, not only in the gateway |
| Pinned addresses | Connect to the exact address that was checked, so DNS can't change in between |
| MFA for password accounts | A second factor for teams that don't use single sign-on |
| SCIM user provisioning (Enterprise) | Add and remove console users from your identity provider |
| Lockdown beyond Docker | Recipes for Kubernetes and cloud VMs, so Blend0 is the only way out everywhere agents run |
| Log rotation | The audit log grows forever today |
Then
Deeper control
| Item | Why |
|---|---|
| Payment decisions in proxy mode | For agents that can't change how they make requests. Needs Blend0 to see inside HTTPS, with your consent |
| Signed audit log | Make changes to the log detectable |
| A message when the advisor acts | A webhook post whenever a change is applied or undone automatically |
Later
Hosted
| Item | Why |
|---|---|
| Hosted Blend0 | The console for many teams, with nothing to run |
| Blend0 Agent | Control which tools and sites agents reach |
?
Open questions
- Do teams need wallet-level enforcement before they'll adopt a gateway?
- Should budgets belong to each agent, or to the team?
- How many recorded outcomes before a reputation score can be trusted?
- Who buys this: platform engineering, security, or finance?
Have an answer? Write to us.
Give your agents a budget, not a blank check.
Blend0 is open source. Ideas and requests are welcome.
contact@acificlabs.com