The system your AI tools run inside.
Seven layers sit between your teams and the AI. Every workflow, prompt, model call and action passes through them — so it's consistent, measured, approved and traceable.
Seven layers. One plain sentence each.
Pick a layer to see what it does — and how it looks inside Sage Runner.
Runner
Workflows are reusable templates — defined once, shared across every project. The Runner executes them step by step, handles retries, and pauses for approval where your policy says so.
Prompt & Skill Store
Every prompt is an organisational asset with an owner, a version history and a model. Improvements are published once and every team gets them; a bad change rolls back in one click.
Critic & Evals
Dedicated critic agents check work against your definitions of done, quality and security before anyone reviews it. Anything risky above your threshold waits for a human.
Memory
Sage keeps a structured picture of each project — services, components, epics and stories — plus learnings confirmed across runs. Agents start from what's known instead of from zero.
Observability
Every run records what went in, which prompt version and model ran, on which engine, what it cost, what came out and who approved it. Any run can be replayed or exported.
Model Rail
Every model call goes through one governed route. Admins hold the keys, choose which models are allowed, and see usage, spend and latency per model — across every provider.
Tool Gateway
Agents never touch your repos or tickets directly. Every action — writing stories to Linear, pushing to GitHub, merging a PR — goes through the gateway, is risk-tiered, and waits for approval when it should.
// risk tiers shown for illustration — [CONFIRM TIERS]
// panels mirror Sage Runner screens · demo data
Each step gets the right agent, engine and model.
This is End-to-End Foundation: a BRD goes in, a full codebase scaffold comes out. Notice the engines change step to step — Sage picks the best tool for each job, and you can swap any of them without touching the workflow.
Open any step. See exactly what happened.
Click a step, then switch between what went in, the prompt version that ran, and what came out. This is what makes AI work auditable — and debuggable.
1Input Assessment Agentassessment · input-assessment/v2.1
2Story Decomposition Agentdecomposition · story-decomp/v1.5
3Story Quality & DoR Agentcritic · story-quality/v1.2
4Linear Writeback Actionaction · connector: linear · risk: LOW
One view of every team, project and dollar.
Runs, success rate, AI cost against budget, what's waiting for approval and what's blocked — across all projects, updated live.
// from the Sage Runner demo · demo data
Runs where your code is allowed to go.
Your code, prompts and accumulated knowledge stay inside your boundary, with the model providers you choose — including self-hosted.
FAQ
Do our developers have to change tools?
No. Developers keep their editor and coding assistants for exploratory work. Sage takes on workflow-level work — requirement to reviewed pull request — and delivers it as stories and PRs where they already work.
Which models and coding agents does Sage use?
The ones you approve. Each workflow step names its engine (Claude Code, Codex, AutoGen and others) and its model, chosen by policy. Your admins control the model registry and keys.
What does a human approve?
Whatever your policy says. Gates can be set per step or by risk score — for example, every story breakdown, or any action scoring 70 or above. Nothing is merged or deployed without a person.
Where does our code and data go?
It stays inside the deployment you choose — managed, your cloud or on-premises. Model calls only go to providers you've approved, through Model Rail.
What happens when a better tool comes out?
We evaluate it on your own workflows, side by side with the current one. If it wins on quality or cost, it's swapped once — centrally — and every team gets it on their next run.
How long does it take to get started?
A pilot starts with one real workflow on your codebase. [PILOT LENGTH] from kickoff to a side-by-side comparison.
See it on your workflow.
A focused pilot on your codebase, run alongside your current tools.