Private preview · Invitation-only managed qualificationRequest access →
Fullbeam
Product

The release platform for coding-agent stacks

Fullbeam gives the whole coding-agent setup a version, covering the model, harness, instructions, tools, permissions, routing, checks, and runtime that shape the result. The current release and candidate then face repeated work from your repos before the new setup earns a single workload.

What the release gate checks

The evidence behind the decision

Each result stays tied to the Stack Release, workload, repository state, and coverage that produced it. If evidence is missing, the gap remains visible in the release call.

01

Give the setup a release

A model name covers one part of the system, so Fullbeam also records the harness, AGENTS.md digest, skills, MCP schemas, permissions, routing policy, verification, and runtime image that can change the result.

02

Run real work more than once

A clean pass can be luck. Use maintenance tasks, migrations, incidents, and follow-up changes from your own repositories, then repeat the cases where a rare miss would hurt.

03

Check what ran

A harness can reroute a model, mount a plugin, or hand work to a subagent after the task starts, so runtime evidence checks whether the accepted change came from the release your team approved.

04

End with a release call

Read acceptance, variance, observed attempt cost, and evidence gaps together. Steering, CI, review, and rework remain separate coverage fields until they are joined to the comparison.

Bring us the next stack change

Promote the candidate where the evidence holds, and keep the current release on workloads that still fail or lack enough coverage to support a wider rollout.

One current releaseOne candidatePrivate repository workA workload-level decision