Changelog
v0.2.0
May 15, 2026
BreakingConfig schema v2
We are migrating from the v1 YAML config format to v2. v1 will be supported until 2026-07-01, after which Maestrio will refuse to start with a v1 config.
What changed
The top-level shape moved from a flat keyed object to a sectioned format:
# v1 (deprecated)default_confidence: suggestbug_confidence: draftdocs_confidence: ship# v2confidence:default: suggestbug: draftdocs: shipWe also renamed the following fields:
human_reviewersis nowreviewers.requiredauto_mergeis nowmerge.strategy, taking valuesmanual,auto, orauto_after_reviewslack_intake_channelsis nowintake.slack.channels
Migrating
Run
maestrio config migratefrom the CLI. It rewrites your config in place and prints a diff. Commit the result and you are done.Why
The flat format started to break down once we added per-category confidence levels. Sections make growth easier without forcing more renames later.
v0.1.0
May 14, 2026
FeatureFirst public release
The first public release of Maestrio is out. Highlights:
- Slack and email intake
- Automatic triage with category, severity, and dedupe
- Linked GitHub PRs for well-scoped fixes
- Per-team confidence levels (suggest, draft, ship)
Sign up at maestrio.ai if you would like early access.
v0.0.8
May 1, 2026
FixRace condition in dedupe, plus a handful of small fixes
A few weeks of accumulated fixes from beta feedback.
Fixed
- Race condition where two near-identical feedback messages arriving within 50ms could both be created as separate parent tickets. Dedupe now serializes per-channel.
- Slack reactions on the original message are now preserved when Maestrio threads a reply.
- Author avatars on the dashboard no longer 404 when an avatar URL contains query parameters.
- Webhook payloads for
pr.openedwere missing theconfidencefield. Restored.
Known still-broken
- Linear sync still drops the
priorityfield for non-bug tickets. Tracked, fix coming next week.
v0.0.5
April 18, 2026
ImprovementFaster Locate, smaller token bills
Locate step now uses a reranker before handing files to the Patch agent. This cuts the average context size we send to the LLM by 4x and improves Patch quality on bugs that touch shared modules.
Behind-the-scenes numbers from a week of production runs:
- Median Locate time: 8.2s, down from 14.7s
- Tokens per Patch call: 6,400, down from 24,100
- Cost per resolved bug: $0.21, down from $0.74
No config change required. Existing runs use the new pipeline automatically.
One regression to flag: very small repos under 200 files sometimes return fewer candidate files than ideal. We are tuning the floor.
v0.0.1
April 2, 2026
FeaturePrivate beta opens
Maestrio enters private beta with a small group of design partners.
What is available:
- Slack intake and triage
- GitHub PR creation at Draft confidence
- Per-team config via YAML
- A very rough dashboard
Known sharp edges:
- Reproduce step is currently English-only.
- No retry queue. Failed runs need to be re-triggered manually.
- The CLI does not exist yet.
If you are reading this and want access, send us a note. We are bringing in a handful of teams per week.