AIppocampus 0.3.3

0.3.3 is a dogfood release for fewer sharp edges.

AIppocampus should feel easier to keep nearby: quieter status, clearer routes, less release-theater, and fewer places where an agent has to guess whether a diagnostic is evidence.

• Recall gets better handholds: source-backed aliases, safer candidate planning, outcome feedback, route labels that survive packet trimming, and an attention-router guard that will not promote a less relevant route just because it scored loudly.
• MCP and public outputs are more careful by default. Search is metadata-only unless snippets are requested, thread ids are bounded, storage dry-runs are public-safe, and sync status gives copyable aippocampus commands.
• Install/update is less fussy. apply --all-local can refresh the local Codex plugin cache when there is one clear target, and status can acknowledge when the foreground agent already sees the tools.
• The test and release path is calmer: focused tests first, pr at most once locally, and release preflight for the few checks that can drift after CI.
• Benchmark and doctor reports are smaller and more honest. A report existing is no longer confused with a public-quality claim.

Still alpha. Still source-backed. Fewer knobs in the doorway.

Compare: v0.3.2...v0.3.3

AIppocampus 0.3.2

0.3.2 is a cleanup release for trust at the edges.

The main change is not louder memory. It is calmer proof. Install, export, maintenance, benchmark, and release paths now say more clearly what happened, what is safe to share, and what still needs a human or operator look.

What Feels Different

• Release checks are less ceremonial. The planner now points to the checks that matter for the changed surface, and local preflight no longer tries to be CI.
• Codex plugin install success is readable at a glance. --json now returns a compact success summary; --operator-json keeps the full deep-debug report. Rollback also has a dry-run preview.
• Public issue attachments have a real metadata-only public-export path. Clean-source text, session ids, anchors, graph labels, and searchable SQLite indexes stay out.
• Search, recall, MCP tool listing, health, and CLI help now have quieter public-safe modes for agents that need the next step, not the whole machine.
• Expected operator failures return structured JSON and useful exit codes instead of tracebacks.
• Maintenance and storage reports are bounded by default, with full audits kept explicit.
• Warm ambient status now separates queued work from actual worker evidence.
• Benchmark reports are more honest about proxies, fixture gates, and public quality claims.

Still alpha. Still source-backed. Less smoke, clearer lantern.

Compare

v0.3.1...v0.3.2

AIppocampus v0.3.1

0.3.1 is a small release about foreground manners.

AIppocampus can now let architecture-native route work be felt without turning the frontstage into a wiring diagram. When attention-router, macro, topology, or local/global cues matter, agents get a tiny navigation hint and the same old boundary: deepen before claims.

Human recall output also stops printing long opaque navigation handles by default. The handle is still there for JSON and MCP callers; people get a short next action instead.

Still alpha. Still source-backed. Less transport noise, more useful handhold.

AIppocampus 0.3.0

0.3.0 is the release where AIppocampus becomes easier to invite into an agent's day.

Continuity still comes from source trails, not from a model pretending it remembers. This release makes the public path calmer: first run, update status, onboard status, sync status, CLI help, recall, AIppo, deepen, explain, and prompt-hook routing now fit together more cleanly.

AIppocampus is still alpha. This release does not claim universal recall quality, hosted-service maturity, all-client coverage, or a stable Python SDK. It gives agents clearer doors to knock on, and clearer reasons to stay quiet.

AIppocampus 0.2.0

0.2.0 is the first broad continuity release after the 0.1.x metadata and discovery repair tags. It is published from tag v0.2.0 at merge commit 451dbb7f.

Highlights

• Reframes AIppocampus around source-backed continuity rather than generic task memory: clean source, source refs, packet trust, and the distinction between navigation scent and evidence now lead the public/docs surface.
• Adds recall packet trust/action grammar: direction_only, reopenable_route, bounded_evidence, source_open, and ignore_or_blocked.
• Advances fresh-thread and ambient recall paths, including action contracts, living cue cache selection, prompt-hook consumption, hot-path recall intent guards, active path packets, warm activation, and the #281 field continuity readout.
• Expands public-safe evidence and benchmark families across Field Continuity, hippocampal recall/hard negatives, multimodal fixtures, E2E50 constraints, semantic robustness, AMemGym, MemoryAgentBench, and claim-boundary maps.
• Adds stronger recall navigation, route diagnostics, repo familiarity, source texture, issue-route quality smoke, and live semantic route actionability readouts.
• Deepens coding-continuity, Dream/Journey, subconscious, sync/storage, encrypted sync, maintenance, package, CI, and release-readiness surfaces.
• Removes more resolved legacy helper shims. Prefer the packaged aippocampus CLI and package-owned module entrypoints.

Compatibility Notes

• Public package metadata now declares 0.2.0.
• Python 3.12 remains the support floor; CI/package metadata cover Python 3.12 and 3.13.
• Runtime still has no mandatory third-party runtime dependencies. Release, development, benchmark, and optional OpenAI Agents extras remain opt-in.
• MCP Registry metadata is aligned to 0.2.0.

Published Surfaces

• PyPI and MCP Registry publication completed through the Publish Python Package And MCP Registry workflow.
• Release workflow run: https://github.com/Sapientropic/AIppocampus/actions/runs/27089329691
• Compare: v0.1.1...v0.2.0

Known Boundaries

Current evidence still does not claim interactive Codex Desktop marketplace click-through, third-party install review, macOS/Linux signed binaries, broad provider/client coverage, production-quality private-history recall, live model quality, or broad superiority claims. Many benchmark and smoke surfaces are contract smokes or diagnostic proxies.