AetherIDE α
Frontier Dark α · deterministic execution loop

An AI agent that edits your codebase, runs your test suites, and fixes its own compiler errors.

Aether indexes your repository into an Abstract Syntax Tree (AST), plans a multi-file change, then runs a closed correction loop against your real tests — re-reading stack traces and patching until the suite is green.

Select a task, then press Tab to authorize the run.
71.4%SWE-bench Verified*
3.2smedian plan latency
1.1MAST nodes / repo
~/repos/payments-svc · agent session Idle
branch step 0 / 0 iterations 0 tokens 0
02 — retrieval

Context Engine: AST-indexed retrieval, not fuzzy grep.

Every symbol resolves to its definition and callers. Hover any file to trace its import graph and see the exact context window Aether would assemble — token budget included, before a single edit is planned.

Workspace
Import graph
Hover a file to trace resolved imports →
Active context
Hover a file to assemble its window.
0 tokens 0% of 200K
context window200,000 cap
Files in window
0
Resolved edges
0
Symbols
0
Retrieval
AST

Aether walks the dependency graph outward from the target symbol, stopping when marginal relevance drops below threshold — so the window stays dense, not padded.

03 — evaluation

SWE-bench comparison matrix.

Sampled on a fixed 300-task internal harness derived from SWE-bench Verified. Toggle the metric to compare resolved-rate accuracy, runtime cost per task, and token efficiency across agents. Best-in-class is highlighted per metric.

Best-in-class Comparison
Model / AgentResolved rateValue
*Illustrative sample figures from an internal harness for demonstration — not an officially published or third-party-verified SWE-bench leaderboard result.
04 — deployment

Seats & self-hosted runners.

Per-seat access to the agent, plus self-hosted runners that execute the deterministic loop inside your own VPC — your source never leaves the perimeter. Size it below; the total recomputes live.

$40 / seat / mo
Standard tier1 – 500 seats
$220 / runner / mo
Each runner = 4 concurrent agent loops0 – 100 runners
6.0M / mo
Scales at 250K tokens per seat + 900K per runner. Overage billed at $0.60 / M.
Estimated total
Billed monthly
$0
per month
Start 14-day trial Talk to solutions eng

Self-hosted runners deploy via Helm into your cluster. Source, ASTs and traces stay inside your VPC; only anonymized eval telemetry is returned.