Skip to main content

Changelog

What changed on main in the last 30 days, one line per commit, newest first. Generated 2026-09-08 at build time from the git history; the API's own version is in the service banner.

curl -sf "$VIGIL_URL/"

2026-09-08​

  • Document plans, exports, VIGIL_PRICE_* and decision rule 17
  • docs.Dockerfile for the vigil-docs service and a CI docs job
  • Snippet runner: every docs page carries one tested bash block
  • Generate the API reference from openapi.json and the changelog from git at build time
  • Docs pages: quickstart, concepts, SDKs, adapters, webhooks, OpenTelemetry, MCP, compliance
  • Docs site scaffold on Docusaurus: config, sidebar, styles
  • Console redirects the old Railway host to app.vigil.supertruth.ai
  • Compliance export bundle with manifest on the ledger
  • Console health route for the Railway health check
  • Calibration report from production (n small; structural finding stands)
  • Deploy safety: Railway health checks, pre-push migration guard, rule 18
  • Lakehouse baseline test accepts the seeded spot truths
  • Wire migration 0010 onto 0009_agent_runs; seed three lakehouse spot truths; lakehouse check derives the warehouse from the env
  • Operations: console domain and monitoring
  • Plans, limits, and monthly statements from the meter
  • Blueprint-drift signal family; one BII definition
  • Operations: role grants and console tables
  • Role grants leave the console's own tables to the console; gate and read-only roles cannot read sessions or keys
  • Lakehouse adversary and ratchet
  • Ignore Serena project directory
  • Detector v2: effect size, change point, fleet FDR; calibration report
  • Receipt check exempts identifiers the run queried by and reads to end of line; refused runs are governed too
  • Remove Studio, Chat and DEMO_MODE from the console; final nav and tour stops
  • Add Potter benchmark runner and README
  • Point rule 7 at the Potter harness report
  • Add synthetic Potter fixtures and the synthetic report
  • Settings page: organization, members and roles from the console database, API keys with scopes, 30-day usage; /admin redirects
  • Add Potter benchmark replay and scoring with tests
  • Upgrade the Anthropic SDK so thinking blocks replay intact in the agent loop
  • Reports: printable system card per agent, incident record per action, human-oversight CSV, control mapping table
  • Agents guide, spec section 4.7 for /ai routes, rules 14 and 15
  • Investigator agent: evidence pack and incident draft for a hold
  • Agent runtime: read-only tools, receipts, streaming, self-governance
  • Operations: lakehouse job, pipeline, warehouse, first-run counts
  • Lab page at /lab: scenario packs, step timeline with gate per step, verdict, links to the ledger records; /fishbowl redirects; reviewers may run packs
  • Alerts page: webhook receivers with one-time secret, event and format selection, signed test ping, delivery log
  • Bronze ingest: no persist on serverless compute
  • Registry page: protected config keys with allowed actors, agent lifecycle set and clear; PUT through the API proxy
  • Databricks scripts derive the SQL warehouse from the service env; Azure warehouse id as fallback
  • Policies page: active policy card, versions table, new version form with live weight sum and ordered thresholds, 30-day re-gate preview
  • Deploy script passes the warehouse id as an argument
  • Deploy script creates the vigil catalog and schemas before registering the pipeline
  • Ledger page: browse by seq range and agent, verify now, JSON and JSONL export, offline verifier instructions
  • Review page at /review: open and closed tabs, keyboard shortcuts, bulk release of identical reasons; /enforcement redirects
  • Lakehouse README: what runs where, deploy, verify, CDF note
  • Lakehouse reconciliation in qa_self and drift guards for the notebooks
  • Databricks job payload, deploy and run-once scripts
  • Silver and gold for VIGIL
  • Agent page: plain-English signal sentences, source and actor columns, ledger entries with chain verify, lifecycle panel, review history
  • Hourly Lakebase to bronze ingest with watermarks
  • Fleet page: live SSE stream, sparklines, last event time, gate and monitoring filters
  • Operations: console sign-in, auth tables, scanner lesson
  • Chain check and watermark query helpers for the lakehouse
  • Magic link confirm page navigates from a click handler; the verify URL is never in the HTML
  • Better Auth magic link with org roles; PIN behind CONSOLE_AUTH=pin
  • Spec 0.4, operations for the webhook worker, rule 10
  • Integrations guide: SDKs, adapters, MCP, and pointers to OTel and webhooks
  • Adapters: Claude Agent SDK hooks, OpenAI Agents tracing processor, LangGraph checkpoint
  • Ingest OpenTelemetry GenAI spans as behavioral events
  • Signed webhooks with retries for gate and review events
  • Console component library on SuperTruth tokens
  • Python and TypeScript SDKs with gate-honoring defaults
  • Policy is a versioned artifact; its digest rides on every score, action, and ledger record
  • Event schema v2: source and actor; authorization and lifecycle registries decide, not the agent
  • MCP server: get_trust, list_holds, explain_score, verify_chain
  • Tests read the audit secret from the environment instead of hardcoding it
  • Docs reflect the completed Lakebase cutover
  • Spec 0.3.1 gate service, operations for vigil-gate, rule 9
  • Add vigil-gate: the only service that can release, reject, or escalate a hold
  • Constrain review_outcome; add hold_expires_at and review_note
  • Record the Lakebase cutover; cutover script tolerates the two expected pg_restore notices
  • Review queue releases, rejects, and escalates through vigil-gate
  • Spec 0.3, rule 8, operations for Alembic, roles and the read-only switch
  • CI migrates the Postgres service from empty and runs the live check on it
  • Lakebase provisioning, role grants migration, cutover script and runbook
  • Add enforcement read routes and the VIGIL_READ_ONLY maintenance switch
  • Meter usage per org per day; rate limit per key
  • Scope every row and the ledger chain to an org
  • Adopt Alembic; baseline migration replaces create_all
  • Add wave 6 agents plan and wave 7 commercial plan
  • Add wave 4 console plan and wave 5 lakehouse plan
  • Add wave 3 plan: independent telemetry and integrations
  • Proxy keeps the upstream path and Authorization header across FastAPI slash redirects
  • Proxy keeps the upstream path and Authorization header across FastAPI slash redirects
  • Add wave 2 plan: tenancy, Lakebase, gate service
  • Add operations runbook
  • Add productization design and wave 0-1 plan; ignore IDE and worktree dirs
  • Spec 0.2: matches the running system
  • Decision rules v1 with adversary self-check
  • Freeze scenario-pack verdicts as regression tests
  • Score through one pipeline; alignment faking samples output length; per-type windows
  • Config-tamper detector accepts the spec payload shape; protected-key list unified
  • Hash-chained HMAC ledger with Postgres append-only triggers, verify and export
  • Replace patent claims in the console footer with a proprietary notice
  • Chat shows real signal breakdown from /scores
  • Add COLLAPSE gate status to the console
  • Add COLLAPSE gate status; composite collapses to zero
  • Console calls the API through a server proxy behind a server-verified PIN; ship public assets
  • Remove history deletion; ledger tables are append-only
  • Enforce bearer API keys with scopes on every route; no-store; CORS allowlist; DB health
  • Remove history deletion from the console; ledger tables are append-only
  • Add pytest harness and CI workflow

2026-08-17​

  • Fix Next 15 async params type in chat/[id] page
  • Point dashboard at stable vigil.supertruth.ai (Railway consolidation)