The system of record for organizational skills
Turn expertise into
governed capability.
Codify expertise as versioned skills, prove they work, govern every release, and keep every AI surface aligned.
- Git-backed
- Eval-driven
- SSO-controlled
- Audit-ready
- RepositoryA commit lands in the skill repository.
- EvaluationThe candidate is scored against its eval suite.
- ApprovalReviewers approve under policy.
- ReleaseThe approved version is promoted and recorded.
- DistributionEvery authorized tool receives the same version.
- FeedbackRun evidence returns to inform the next change.
Example workspace
Illustrative figures from the Savant demo workspace- Skills governed
- 218+
- +14 this quarter
- Eval coverage
- 94%
- +2.1 pts / 30d
- First-pass acceptance
- 81%
- +0.4 pts / 30d
- Release turnaround
- 2.4d
- −0.6d vs prior month
Most teams ship skills on trust.
Savant ships them on evidence.
Expertise is becoming executable through AI, but most organizations have no control over it. Prompts, runbooks, and agent instructions drift across tools with no version, no measurement, and no trail when something regresses.
Skills scattered across
- repos
- documents
- chat threads
- local machines
- agent configs
- wikis
- no shared baseline
- no evaluation
- no release governance
- no clear rollback
One governed lifecycle
- 01Source
- 02Evaluate
- 03Approve
- 04Release
- 05Distribute
- 06Learn
See the workspace your team actually governs in.
Repository sync, evaluation outcomes, approval queues, release state, and audit evidence in one view. This is the product, recreated with demo data.
| Skill | Δ quality | Version | State | When |
|---|---|---|---|---|
| Architecture Review | +4.8 | v2.5.0 | Candidate | 4m |
| Incident Triage | — | v3.1.0 | Regression detected | 26m |
| Proposal Response | +3.1 | v1.9.0 | Staged | 1h |
| Engineering RFC Reviewer | +2.1 | v4.0.2 | Production | 3h |
- 01
Health from evidence
Skill health is the eval pass rate, not a status someone typed.
- 02
Approvals under policy
Reviews route by tier. The queue shows who is blocking and for how long.
- 03
Regressions before release
A failing candidate is held automatically and flagged for its owner.
Expertise becomes infrastructure.
Savant turns organizational methods, standards, and expert workflows into measurable capabilities that can be reused across people and AI. Each tier inherits from the one above it, so a change to a standard reaches every skill that depends on it.
- Tier 1
Standards
Organization-wide rules every skill inherits.
- Writing
- Governance
- Security
- Brand
- Evidence
inherited by - Tier 2
Expertise
Domain methods owned by the teams who practise them.
- Architecture
- Sales
- Research
- Engineering
- Operations
specialized into - Tier 3
Workflows
How work actually gets done, for a person, team, or customer.
- Personal
- Team
- Role
- Customer
- Context
Every use can make
the skill better.
Real runs generate evidence. Savant measures outcomes, proposes bounded improvements, and validates each candidate against the regression suite. Only people approve, and nothing reaches production outside your release policy.
- Baseline · v2.4.0
- 87.4
- Candidate · v2.5.0
- 92.6
- Δ quality
- +5.2
Observed behavior
18% of runs required manual trade-off analysis edits.
Proposed improvement
Require operational complexity analysis for each alternative.
Measured impact
- Quality
- +5.2
- Edit rate
- −6.5%
- Latency
- +1.8%
Illustrative example. Approval always stays with your reviewers.
Six stages. Every event recorded.
From the commit that wrote a skill to the tool that runs it, each state is observable and each transition leaves evidence.
- 01
Connect
Point Savant at the Git repository where skills live. Webhook sync keeps every change visible.
acme/skills · main - 02
Evaluate
Each candidate runs against rubric evals and its regression suite before review opens.
148 cases · 24s - 03
Approve
Owners, reviewers, and compliance act on one timeline, routed by tier and policy.
2 of 2 approved - 04
Release
Promote draft → staging → production. Every release is signed and pinned to a commit.
v2.5.0 · signed - 05
Distribute
The approved version reaches every authorized tool through integrations or the sync agent.
4 surfaces in sync - 06
Audit
Every event is recorded append-only, exportable to your SIEM, with one-click rollback.
append-only
Approved skills reach every tool as one version.
Savant keeps the canonical release and propagates it outward: native integrations where tools support remote delivery, a managed sync agent for developer environments.
- ClaudeNative integrationv2.5.0in sync
- OpenAINative integrationv2.5.0in sync
- CodexManaged sync agentv2.5.0in sync
- GitHub CopilotManaged sync agentv2.5.0in sync
- VS CodeManaged sync agentv2.5.0in sync
- One approved version.
- Distribution starts from the release your reviewers approved, never a local copy.
- Every authorized surface.
- Policy decides which tools and groups receive a skill before anything ships.
- No silent drift.
- Each surface reports the version it runs, so a mismatch is visible and one rollback fixes all of them.
Governance isn't an enterprise add-on.
It's the product.
Identity, permissions, provenance, policy, audit, and rollback are part of every plan, because a skill you can't trace is a skill you can't ship.
Identity
- SSO — OIDC and SAMLAuth0 by default, or bring your own identity provider.
- SCIM provisioningMembership and groups stay in lockstep with your directory.
- RBAC + policy controlsRoles and approval tiers codified in policy, not chat.
Provenance
- Git provenanceEvery version resolves to a commit in your own repository.
- Signed releasesRelease records are signed and pinned to evaluated content.
- Immutable audit historyAppend-only events for every change, approval, and rollout.
Operations
- RollbackAuto-pin on regression; one-click return to a prior release.
- SIEM exportStream audit events to the tools your security team runs.
- Tenant isolationEach workspace is isolated, and skill content stays in your Git.
One platform.
One plan.
Everything required to govern organizational skills.
- Unlimited skills
- Evaluations
- Releases
- Repositories
- SSO + SCIM
- Distribution
- Audit
- Improvement recommendations
What teams ask before they connect a repository.
Something else? Email hello@savant.app. We reply within a day.
Does Savant store our skill content?
No. Your Git repository is the source of truth. Savant references commits and runs evaluations against them, but prompts, runbooks, and agent workflows stay in your environment.
Which Git providers do you support?
GitHub Cloud and Enterprise, GitLab Cloud and self-managed, Azure DevOps, Bitbucket Cloud and Data Center, and other Git deployments over SSH or HTTPS.
What does an eval suite look like?
A rubric and a case set, both checked into the repository. Savant scores each candidate with the model you choose, surfaces regressions against the baseline, and keeps results for the life of the release.
Does Savant change skills on its own?
No. Savant proposes improvements from run evidence and validates each candidate against the regression suite, but only authorized reviewers approve. An approved change becomes a new Git version and follows your existing release policy.
How does authentication work?
Auth0 by default, or bring your own identity provider over SAML or OIDC. Group membership drives RBAC, and SCIM keeps it in lockstep with your directory.
What happens if a release regresses?
Auto-pin on regression is a default policy: the prior version pins immediately, an incident opens, and the skill owner is notified. Manual rollback is one click.
Do you offer a free trial?
Yes. Every workspace starts with a 14-day trial with all workflows enabled. Cancel during the trial and you are not charged; after it, billing is per seat, monthly or annually.
Stop shipping expertise
on trust.
Evaluate it. Govern it. Improve it.
Put your organization's expertise under governance. Connect a repository, watch the first evaluation run, and ship a signed release in under fifteen minutes.