Skip to content

Releases: sysroot-io/akritas

Release list

0.2.0

0.2.0 Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 09 Sep 12:39
0.2.0
92d64d9

Full Changelog: 0.1.0...0.2.0

Assets 8
Loading

0.1.0

0.1.0

Choose a tag to compare

View all tags
@Volant Volant released this 01 Sep 00:43
0.1.0
67ae939

Added

  • Added a bundled read-only VictoriaMetrics MCP server.

  • Added instant and range MetricsQL/PromQL queries.

  • Added bounded series, label, and label-value discovery tools.

  • Added example VictoriaMetrics MCP configuration.

  • Added a tool catalog dialog to the Web UI.

  • Added host-owned Run budgets covering:

    • Model iterations and output tokens.

    • Submitted context size.

    • Tool calls and result size.

    • Retrieved context size.

    • Total Run duration.

  • Added Run budget usage to Chat responses and the Web UI.

  • Added persistent investigation results to the append-only audit store.

  • Added structured start and finish logs for tool execution, including duration and status.

  • Added tool execution duration to Web UI activity entries.

  • Added Markdown table and horizontal-rule rendering to model responses.

Changed

  • Replaced local.capability.report_missing with local.investigation.submit_plan.

  • Investigation planning now produces both executable checks and unavailable capability gaps.

  • Akritas now supplies the planner with authorized read-only tool names, descriptions, and input schemas.

  • The host validates and executes every accepted planned check before final model reasoning.

  • Denied, unavailable, write, and dangerous tools are excluded from investigation planning.

  • Runbook content is treated strictly as guidance and cannot create or authorize capabilities.

  • OpenAI-compatible model calls now participate in Run budget accounting.

  • Missing upstream token usage is charged conservatively using the reserved allowance.

  • Alertmanager responses now include structured investigation and budget objects.

  • Chat responses now include Run budget usage.

  • Health and tool catalog counts now reflect authorized tools.

  • Audit metadata now records bounded Run usage counters.

  • Failed chat requests restore the submitted message in the Web UI for retry.

  • Change proposal diagnostics now support downloading the complete structured proposal.

Security

  • Added strict validation of planned tools, arguments, limits, and duplicate steps.

  • Added host validation of all structured investigation evidence references.

  • Prevented model confidence from affecting authorization decisions.

  • Added bounded VictoriaMetrics query, matcher, discovery, response, and error-preview limits.

  • Kept VictoriaMetrics endpoints and credentials outside model-controlled arguments.

  • Added explicit documentation for VictoriaMetrics scope, debug-log, and expensive-query risks.

  • Expanded the threat model for fabricated evidence, resource exhaustion, and metrics scope escape.

User Interface

  • Added a detailed Run log with Run ID, tool status, execution duration, and budget usage.

  • Added separate presentation of executed tools and unavailable checks.

  • Added an authorized tool catalog accessible from the header.

  • Improved change preparation and approval diagnostics.

  • Added labels, placeholders, status messages, errors, dialogs, and accessibility text.

Documentation

  • Documented investigation planning and host-executed checks.

  • Documented Run budget configuration and accounting.

  • Documented the VictoriaMetrics MCP adapter and troubleshooting mode.

  • Updated architecture, security model, threat model, README, and operations guide.

Loading