v0.9.3 — Load, e2e, ops polish

Meridian v0.9.3

Load/overhead numbers, richer e2e, and multi-tenant ops polish on the 0.9.x track (v1.0 still deferred).

Highlights

• Load — scripts/bench_overhead.py + docs/LOAD.md (~0.7 ms serial p50 overhead on mock path)
• E2E — enterprise suite (auth, budgets, cost, stream, failover); CI gateway-smoke (live mock + Meridian + smoke)
• Ops — x-meridian-budget-remaining-tokens|requests; docs/OPS_RUNBOOK.md; retention/sizing in DEPLOY

Images

• ghcr.io/imv-in/meridian:0.9.3 / :latest (release workflow)

Validate

python scripts/bench_overhead.py --requests 100 --concurrency 1
python scripts/smoke_test.py --url http://localhost:8080
pytest -q

Next

Optional multi-provider / semantic cache / batch, or design-partner v1.0 gate when product proof is ready.

v0.9.2 — Budget ↔ actual reconcile

Meridian v0.9.2

Budget ↔ actual token-meter reconcile on the 0.9.x track (v1.0 still deferred).

Highlights

• Pre-flight budgets still reserve on estimate
• After successful response with backend usage, token meters adjust to actual cost (prompt * prefill_weight + completion * decode_weight)
• Request counters never adjusted; failed/upstream-error requests keep the estimate
• Usage scrape when cost or budgets need tokens
• Metric: meridian_budget_reconciles_total{direction}

Images

• ghcr.io/imv-in/meridian:0.9.2 / :latest (release workflow)

Next (0.9.3)

Load/overhead numbers, richer e2e, ops polish for multi-tenant fleets.

v0.9.1 — Product-complete polish

Meridian v0.9.1

Product-completion polish on the 0.9.x track (v1.0 still deferred).

Highlights

• `GET /meridian/version` for ops smoke
• `configs/enterprise_example.yaml` production-shaped template
• `cost.enabled` requires `auth.enabled` at process start
• Helm chart 0.9.1 + CI `helm template` job
• Docs: ship/ROADMAP/MILESTONES — continue 0.9.x before any 1.0 gate

Images

• `ghcr.io/imv-in/meridian:0.9.1` / `:latest` (release workflow)

Next (0.9.2)

Budget ↔ actual reconcile, load numbers, further enterprise polish.

v0.9.0 — Deployment packaging (Milestone N)

Meridian v0.9.0

Milestone N — deployment packaging for enterprise PoCs and air-gapped installs.

Highlights

• Helm chart: deploy/helm/meridian/
• Air-gap: scripts/package_airgap.sh + docs/AIRGAP.md
• Ops: docs/DEPLOY.md (TLS, backup, sizing, key rotation)
• Key hot-reload: auth.keys_file + SIGHUP or POST /meridian/reload (ops_admin)
• Full history: docs/MILESTONES.md

Images

• ghcr.io/imv-in/meridian:0.9.0 / :latest (via release workflow)

Not 1.0 yet

Product completion continues on 0.9.x. v1.0 remains a later verification gate after the product is complete for design partners.

v0.8.0 — Cost attribution (Milestone M)

Meridian v0.8.0

Milestone M — cost attribution with enterprise hardening.

Highlights

• Opt-in cost: — actual usage from non-stream JSON and stream SSE (last wins)
• Ledger: memory or sqlite (WAL); per-model $/1M prompt & completion
• GET /meridian/usage + /meridian/usage.csv
• Auth required when cost is on; non-admin org/team scoped; cost_admin for all orgs
• Metrics: meridian_tokens_total{model,kind}
• Checklist: docs/ENTERPRISE_COST.md

Images

Release workflow publishes:

• ghcr.io/imv-in/meridian:0.8.0 / :latest
• Docker Hub when configured: lothnic0801/meridian:0.8.0

Enable carefully

auth: { enabled: true, keys: [...] }
cost:
  enabled: true
  store: sqlite
  sqlite_path: /var/lib/meridian/meridian_cost.db

Do not enable cost with auth.enabled: false (usage API returns 401).

Next

Milestone N — Helm, air-gap bundle, key reload.

v0.7.0 — PII India pack (Milestone L)

Meridian v0.7.0

Milestone L — India PII detection & redaction, on top of the full A–K stack.

Highlights

• Opt-in request-path PII pack: Aadhaar (Verhoeff), PAN, GSTIN, IFSC, UPI, Indian mobile
• Policies: block · redact_and_replace · redact_for_logs · audit_only
• Per-key pii_policy override
• Matched values never logged (counts by entity type only)
• Metrics: meridian_pii_detections_total{entity,policy}

Also in this line (already on main / prior tags)

• Identity: auth, model access, org rate limits, tenant budgets (v0.4–v0.5)
• Hardening: bounded rate-limit store, stream-safe cleanup, body cap, non-root image (v0.6.0)
• Routing: token-aware, tiering, session affinity, telemetry

Images

Publish via release workflow (GHCR / Docker Hub when configured):

• lothnic0801/meridian:0.7.0 / :latest
• ghcr.io/imv-in/meridian:0.7.0 / :latest

Validate

docker pull lothnic0801/meridian:0.7.0   # or ghcr mirror
pytest -q   # from source: 278+ tests

Next

Milestone M — Cost attribution (actual usage, price tables, /meridian/usage).

Full notes: CHANGELOG.md · ship.md