Skip to the demo

Symbol graph · clean merges that break callers

Clean merge. Broken callers · Caught — watch the difference.

An agent changes calculate(a,b) from a + b to a * b. Git has nothing to fight about — three dependents still break. Toggle Without Managent (silent breakage) and With Managent (Semantic Impact traces, validates, and can block with evidence). Opt-in per repo; not a claim of perfect behavioral understanding. Numbers here are demo/replay evidence.

With Managent · semantic impact

Clean merge. Conflict caught.

Symbol graph finds the callers. Validation confirms the break. Enforce blocks only with evidence — Observe and Warn stay quieter by default.

With Managent: an agent changes calculate from addition to multiplication. Git merge is clean. Semantic Impact finds three dependents, validation confirms the break, and Enforce blocks the merge with evidence — not AI suspicion alone.

semantic impact: waiting for analysis…

What just happened, mechanically

1

Change

An agent edits a shared symbol — here calculate(a,b) flips from addition to multiplication. Paths may not overlap with callers, so git merge-tree stays clean.

src/math/file2.ts · calculate · return a + b → a * b

2

Trace

Semantic Impact walks the symbol graph and finds bounded dependents — calculateOrderTotal, calculateInvoice, buildRevenueReport — then scores the risk.

POST /v1/semantic-analysis → calculate → 3 dependents · risk 88

3

Validate

Suspicion alone never blocks. Targeted tests or typecheck must confirm incompatibility before a risk becomes a semantic conflict.

validation · FAIL expected 15 got 50 · confirmed

4

Act

Observe stays on the dashboard. Warn annotates the PR. Enforce blocks only confirmed conflicts above threshold — opt-in per repository.

mergeMode=enforce · GitHub Check · Semantic conflict — merge blocked

Honest edition: Semantic Impact is opt-in and separate from territory leases. Published textual conflict rates stay merge-replay only — semantic signals never enter the lease allocator. Enforce blocks confirmed conflicts with deterministic or test evidence; AI-only suspicion never blocks a merge.

Run it on your repo

Enable Semantic Impact in the workspace, pick Observe / Warn / Enforce, and let agents sync awareness over MCP when a symbol change hits dependents.

# opt-in · dashboard → Semantic impact → Warn

$PUT /v1/semantic-policies · enabled, mergeMode: warn

# agents sync awareness over MCP

$claude mcp add managent -e MANAGENT_API_KEY=mgt_live_… -- managent mcp

Managent is invite-only during early access and free while we onboard design partners. Patent pending.

We onboard a few fleets each week, in order of request.