- Model-neutral
- Frontier models where judgment matters, open-weight models where they’re good enough. Switch any role to any model.
- Workspace-native
- A persistent cloud workspace per developer, with live preview URLs. Agents and people work in the same place.
- Secure by construction
- Each agent session sees only the repos it’s working on. Agent sandboxes hold no platform, organization or cloud credentials.
- Opinionated method
- A complete agentic SDLC: plan, build, independent review, merge, and learn. Built in, not bolted on.
One run, a team of agents
From ticket to merged PR, with every step visible.
The Orchestrator is the agent you talk to. It delegates to a Planner, Builders, Explorers and an independent Reviewer, each in its own isolated session with the model and reasoning effort that fits the job.
Run · Add retry budget to webhook delivery
acme/payments-api · Standard lane · autonomy A3
- Orchestrator
Picked up PROJ-482 from the tracker
open-weighteffort: low - Explorer ×3
Mapped the affected modules (read-only)
smalleffort: none - Planner
Wrote the plan, 3 slices; approved
frontiereffort: high - Builder
Slice 2 of 3: mise run verify passing
open-weighteffort: medium - Reviewer
Independent review on a different model family
frontiereffort: high
Illustrative. Each role runs as its own isolated session with its own model and reasoning effort.
Platform
Built for enterprise engineering, not demos.
Any model
Right model, right effort, every role
Route planning and review to frontier models, and routine building and exploration to open-weight models at a fraction of the cost. Bring your own OpenRouter account; managed access follows after launch.
Isolation
Agents see only their repos
Every agent session runs in its own sandbox containing just the repositories it was given. Agent sandboxes hold no platform, organization or cloud credentials.
Workspaces
Run it, see it, ship it
Persistent cloud workspaces with authenticated preview URLs for any port. They sleep when idle and wake in seconds.
Merge gate
This run read an issue from an outside contributor.
- Checks passing
- Reviewer approved (different model)
- One human approval required
Oversight
Humans where it matters
Agents merge on their own within your policy. Runs that touched untrusted outside content wait for a person.
Deployment
Our cloud or yours
Multi-tenant or dedicated SaaS, managed in your own cloud, or a licensed self-hosted edition that runs fully offline.
The Portfall Loop
A method that gets agent PRs merged.
Plan first
Separate thinking from typing
Explore, plan and approve before a line is written. Plans are small, testable slices with a single command that proves health.
Integrations
Meet your team where it works
Planned: GitHub, GitLab and Forgejo first, then Bitbucket and Azure DevOps; Jira, Linear and Notion; and any MCP server.
Review
An independent second opinion
The Reviewer never grades its own work: it runs on a different model family and reviews the diff against the plan and the evidence, in a fresh, read-only context.
Compound
Every merge makes the next one easier
After each merge, lessons become repository knowledge: solution notes, rules and tests that future runs start from.
Adoption services
The tools are the easy part.
Most teams don’t struggle to try coding agents; they struggle to turn them into a delivery system that holds up under review, security and scale. We help engineering leaders make that transition, with or without our platform.