tcconnally/perseus](https:

v1.0.24 Feature

This release adds 2 notable features for engineering teams evaluating rollout.

Published 10d AI Agents & Assistants

✓ No known CVEs patched

✓ No known CVEs patched in this version

Topics

agent-memory ai-agents claude-code cli context-engine context-engineering

+9 more

hermes llm local-first mcp mcp-server model-context-protocol perseus python workspace-context

Summary

AI summary

New honest prompt-token A/B harness reduces tokens by 52.63% and adds savings‑wire metering integration with Plutus 1.1.0.

Type	Severity	Summary	CVE
Feature
Feature	Medium	Added honest prompt-token A/B harness reducing prompt tokens by 52.63%. Added honest prompt-token A/B harness reducing prompt tokens by 52.63%. Source: llm_adapter@2026-07-16 Confidence: high	—
Feature	Medium	Implemented savings wire metering to Plutus (v1.1.0) with counterfactual baselines. Implemented savings wire metering to Plutus (v1.1.0) with counterfactual baselines. Source: llm_adapter@2026-07-16 Confidence: high	—
Feature	Low	Added meter_context_reduction() helper for memory-posture reduction estimates. Added meter_context_reduction() helper for memory-posture reduction estimates. Source: granite4.1:30b@2026-07-16-audit Confidence: high	—
Performance	Low	Render overhead p50 is 341.5 ms cold / 315.6 ms warm inside the window. Render overhead p50 is 341.5 ms cold / 315.6 ms warm inside the window. Source: granite4.1:30b@2026-07-16-audit Confidence: high	—
Deprecation	Medium	Retired synthetic 94% / 0 ms / 1,190x headline from all public surfaces. Retired synthetic 94% / 0 ms / 1,190x headline from all public surfaces. Source: llm_adapter@2026-07-16 Confidence: high	—
Bugfix	Low	Enforced perseus_tool_count via test and registered reduction figure as derived claim. Enforced perseus_tool_count via test and registered reduction figure as derived claim. Source: llm_adapter@2026-07-16 Confidence: high	—

Full changelog

Claims-integrity and savings-wire release.

Retired the synthetic 94% / 0 ms / 1,190x headline from every public surface; claims.json records why, CI bans the retired figures, and the site calculator/demo are rebased on the cost-savings certification with the full-context baseline labeled as an upper bound (#803).
New honest prompt-token A/B harness with a measured, committed artifact: 52.63% fewer prompt tokens (naive assembly vs shipped defaults on the Perseus repo corpus, full-request tiktoken counting, symmetric arms, no penalties; includes a no-windowing control at -8.69%), render overhead p50 341.5 ms cold / 315.6 ms warm timed inside the window. benchmark/tokenab/report.json, sha256 content-hashed (#804).
Savings wire: meter_response/meter_usage forward counterfactual baselines to Plutus (plutus 1.1.0 / #134); new meter_context_reduction() helper; opt-in plutus.meter_memory_posture records per-render memory-posture reduction estimates into a dedicated estimates workspace. WIRING.md section 9 (#805).
perseus_tool_count is now genuinely code-enforced by test; the tool-schema reduction figure is a registered derived claim; README infographic re-anchored to artifact-backed figures (#807).

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Track tcconnally/perseus](https:

Get notified when new releases ship.

About tcconnally/perseus](https: