diff --git a/ARCHITECTURE.md b/ARCHITECTURE.md
index 29ce2fd..a43eb40 100644
--- a/ARCHITECTURE.md
+++ b/ARCHITECTURE.md
@@ -58,7 +58,7 @@ work around it. Stability of the public surface is governed by
 | `gateway` | Surface 1 — OpenAI/Anthropic-compatible proxy; meters + governs every call. |
 | `mcp` | MCP gateway — intercepts `tools/call` for allowlist + approval + audit. |
 | `memory` | Git-native memory reader (user-owned md/yaml; path-traversal-safe; keyword search). |
-| `otel` | Surface 3 — OpenTelemetry GenAI span export. |
+| `otel` | Surface 3 — OpenTelemetry GenAI span export + an OTLP/HTTP trace ingress (`POST /v1/traces`) that meters consumed spans. |
 | `httpapi` | HTTP server: mounts the proxy, the `/v1` run-control API, the memory + approval endpoints, and the local admin page. |
 | `config` | Config from env + `.env`. Secrets only from here; never stored/logged. |
 | `httpx`, `id`, `version`, `app` | Small shared helpers: JSON responses, UUIDs, build identity, bootstrap wiring. |
@@ -96,5 +96,5 @@ work around it. Stability of the public surface is governed by
 SQLite (WAL) is the default store — one file the user owns. Tables: `runs`,
 `steps`, `tool_calls`, `cost_ledger`, `checkpoints`, `approvals`, `memory_facts`.
 Migrations are embedded and **forward-only** (the daemon refuses to start if the
-on-disk schema is newer than the binary). Postgres is a future opt-in behind the
-same `Store` interface.
+on-disk schema is newer than the binary). Postgres is an opt-in backend behind the
+same `Store` interface (`RISKKERNEL_DATABASE_URL`), for multi-instance / HA.
diff --git a/README.md b/README.md
index 8180925..10b1935 100644
--- a/README.md
+++ b/README.md
@@ -37,15 +37,17 @@ It is **not** another gateway (LiteLLM/Portkey own routing), **not** another obs
 | 🔁 **Hard loop-iteration cap** | No more infinite agent loops. |
 | ⏱️ **Hard wall-clock budget** | Runs that exceed their time budget halt. |
 | 💾 **Crash-resumable checkpoints** | `kill -9` the daemon mid-run; it reloads with the budget already spent and resumes from the last checkpoint — without re-spending. [Guide](docs/RESUME.md) · [demo](examples/kill-9-resume). |
-| ✋ **Framework-agnostic approval gates** | Side-effecting tool calls pause for human approval — CLI, local web, or webhook. |
-| 🧠 **Memory you own** | Git-native markdown/YAML on your disk; episodic state in your SQLite. |
-| 📡 **OpenTelemetry GenAI** | Emits `gen_ai.*` spans to *your* backend (Grafana/SigNoz/Datadog/Langfuse). |
+| ✋ **Framework-agnostic approval gates** | Side-effecting tool calls pause for human approval — CLI, local web, webhook, or **Slack**. |
+| 📜 **Policy-as-code** | Reusable budget / tool-allowlist / approval bundles via `POST /v1/policies` or a reviewed `riskkernel.yaml`, dry-run against recorded runs ([the policy guide](docs/POLICY.md)). |
+| 📊 **Spend attribution & compliance** | Roll cost up across runs by team/user/feature (`riskkernel audit summary --by metadata.team`), plus a tamper-evident OWASP / EU AI Act evidence export ([compliance](docs/COMPLIANCE.md)). |
+| 🧠 **Memory you own** | Git-native markdown/YAML on your disk; episodic state in your SQLite (or opt-in Postgres for HA — [docs](docs/POSTGRES.md)). |
+| 📡 **OpenTelemetry GenAI (both ways)** | Emits `gen_ai.*` spans to *your* backend (Grafana/SigNoz/Datadog/Langfuse) **and ingests** them, to meter apps it never proxied ([ingress](docs/OTLP_INGRESS.md)). |
 
 ## Three ways to adopt — pick the one that fits
 
-1. **Proxy (zero code).** Set one env var: `OPENAI_BASE_URL=http://localhost:7070/v1`. Every call is intercepted, budgeted, logged, checkpointed, and forwarded to the real provider with your key.
-2. **Python SDK (deep control).** Install the SDK (from source today — see the [Quickstart](#quickstart-60-seconds)), then `@governed_run` / `@governed_tool` / `runtime.budget(...)` / `ApprovalGate`. Adapters for the Claude Agent SDK, OpenAI Agents SDK, and LangChain.
-3. **OpenTelemetry (universal).** RiskKernel is an OTLP endpoint *and* emitter — govern apps already instrumented with OpenLLMetry / the OpenAI Agents SDK, and export to the backend you already run.
+1. **Proxy (zero code).** Set one env var: `OPENAI_BASE_URL=http://localhost:7070/v1` (or `ANTHROPIC_BASE_URL` for `/v1/messages`). Every call — streaming or not — is intercepted, budgeted, logged, checkpointed, and forwarded to the real provider with your key. Native providers: Anthropic, OpenAI, and Ollama (local).
+2. **SDK (deep control).** `pip install riskkernel` (Python) or `npm install @riskkernel/sdk` (TypeScript), then governed runs, per-step loop/time budgets, checkpoints, and approval gates. Framework adapters for the Claude Agent SDK, OpenAI Agents SDK, LangChain, LlamaIndex, CrewAI, AutoGen, and PydanticAI (Python), and the Vercel AI SDK (TypeScript).
+3. **OpenTelemetry (universal).** RiskKernel is an OTLP endpoint *and* emitter — ingest GenAI spans (`POST /v1/traces`) to meter apps already instrumented with OpenLLMetry / the OpenAI Agents SDK / the Vercel AI SDK, and export cost/halt/tool spans to the backend you already run.
 
 ## Quickstart (60 seconds)
 
@@ -87,9 +89,11 @@ curl -s -D- http://localhost:7070/v1/chat/completions \
 Inspect and audit, all on your disk:
 
 ```bash
-riskkernel runs list                 # every governed run
-riskkernel audit export <run-id>     # the cost ledger as JSON
-riskkernel audit tools <run-id>      # governed tool calls as JSON
+riskkernel runs list                      # every governed run
+riskkernel audit export <run-id>          # the cost ledger as JSON
+riskkernel audit tools <run-id>           # governed tool calls as JSON
+riskkernel audit summary --by metadata.team   # spend rolled up across runs
+riskkernel audit compliance <run-id>      # OWASP / EU AI Act evidence export
 ```
 
 Prefer a native binary to Docker? Install the CLI with one command — no clone
@@ -109,14 +113,16 @@ riskkernel completion zsh  > "${fpath[1]}/_riskkernel"                # zsh
 riskkernel completion fish > ~/.config/fish/completions/riskkernel.fish  # fish
 ```
 
-Deeper control (loops, checkpoints, approval gates) is the Python SDK:
+Deeper control (loops, checkpoints, approval gates) is the SDK — Python or
+TypeScript:
 
 ```bash
-pip install riskkernel
+pip install riskkernel          # Python  → sdks/python
+npm install @riskkernel/sdk     # TypeScript → sdks/typescript
 ```
 
-See [`sdks/python`](sdks/python). Trace every run in your own backend:
-[`examples/otel`](examples/otel).
+See [`sdks/python`](sdks/python) and [`sdks/typescript`](sdks/typescript). Trace
+every run in your own backend: [`examples/otel`](examples/otel).
 
 Want to *see* the headline feature? [`examples/codebase-qa`](examples/codebase-qa)
 is a runnable agent that loops over a codebase until the governor kills it on its
diff --git a/ROADMAP.md b/ROADMAP.md
index f807a74..4a1af38 100644
--- a/ROADMAP.md
+++ b/ROADMAP.md
@@ -16,46 +16,47 @@ The core runtime is built and released:
   that halt a run *before* it overspends ([`docs/BUDGETS.md`](docs/BUDGETS.md)).
 - **Crash-resume** — `kill -9` a run mid-flight and resume it without re-spending
   ([`docs/RESUME.md`](docs/RESUME.md)).
-- **OpenAI-compatible proxy** — point one env var at RiskKernel and every call is
-  metered, priced, and budget-enforced (BYO key).
+- **OpenAI- and Anthropic-compatible proxy** — point one env var at RiskKernel and
+  every call (streaming or not) is metered, priced, and budget-enforced (BYO key).
+  Native providers: Anthropic, OpenAI, and Ollama (local, key-free).
 - **Human-in-the-loop approval** — gate side-effecting tools; resolve from the CLI,
   a local web page, a webhook, or **Slack** ([`docs/APPROVALS_SLACK.md`](docs/APPROVALS_SLACK.md)).
-- **Policy-as-code** — named policy bundles via `POST /v1/policies` or a reviewed
-  `riskkernel.yaml`, with a dry-run against recorded runs ([`docs/POLICY.md`](docs/POLICY.md)).
-- **OpenTelemetry GenAI export** — cost/halt/tool spans into your existing backend,
-  with a ready-made Grafana + Tempo dashboard.
+- **Policy-as-code, enforced per-run** — named policy bundles via `POST /v1/policies`
+  or a reviewed `riskkernel.yaml`, with a dry-run against recorded runs; a run created
+  under a bundle is governed by its tool allowlist and approval rules, not just its
+  budget ([`docs/POLICY.md`](docs/POLICY.md)).
+- **OpenTelemetry GenAI — export and ingress** — emit cost/halt/tool spans into your
+  existing backend (ready-made **Grafana + Tempo** and **SigNoz** dashboards), *and*
+  ingest GenAI spans (`POST /v1/traces`) to meter apps RiskKernel never proxied
+  ([`docs/OTLP_INGRESS.md`](docs/OTLP_INGRESS.md)).
+- **Spend attribution** — roll cost up across runs by team/user/feature
+  (`riskkernel audit summary --by metadata.team`), with the run name and tags also on
+  the OTel spans so the same grouping works in your backend.
 - **MCP gateway** — govern an agent's `tools/call` (allowlist + approval + audit).
 - **Compliance evidence export** — controls mapped to OWASP / EU AI Act references
   with a tamper-evident, hash-chained event log ([`docs/COMPLIANCE.md`](docs/COMPLIANCE.md)).
 - **Git-native memory** — markdown/YAML the agent reads, that you own.
+- **Storage** — zero-config SQLite by default; an opt-in **Postgres** backend for
+  multi-instance / HA behind the same `Store` interface ([`docs/POSTGRES.md`](docs/POSTGRES.md)).
 - **SDKs** — Python (`pip install riskkernel`) and TypeScript (`@riskkernel/sdk`),
-  with LangChain, OpenAI Agents, and Vercel AI SDK adapters.
+  with adapters for the Claude Agent SDK, OpenAI Agents, LangChain, LlamaIndex,
+  CrewAI, AutoGen, and PydanticAI (Python), and the Vercel AI SDK (TypeScript).
+- **Operability** — a Prometheus `/metrics` endpoint, a `riskkernel doctor` setup
+  check, shell completions, and a one-command `docker compose` quickstart.
 - **Measured, low overhead** — the enforcement decision is ~150 ns and zero
-  allocations per call ([`docs/PERFORMANCE.md`](docs/PERFORMANCE.md)).
+  allocations per call, with a reproducible cost benchmark and a timed `kill -9`
+  recovery benchmark (exact-once spend) behind the claims ([`docs/PERFORMANCE.md`](docs/PERFORMANCE.md)).
 
 ## Next
 
 Where the work is heading near-term:
 
-- **Per-run policy enforcement** — a referenced policy bundle applies its budget to a
-  run today; extend that to enforce the bundle's tool allowlist and approval rules
-  per-run, not just globally.
-- **More framework adapters** — CrewAI ([#83](https://github.com/prashar32/riskkernel/issues/83)),
-  AutoGen ([#84](https://github.com/prashar32/riskkernel/issues/84)),
-  PydanticAI ([#85](https://github.com/prashar32/riskkernel/issues/85)),
-  LlamaIndex ([#86](https://github.com/prashar32/riskkernel/issues/86)).
-- **More native providers** — AWS Bedrock ([#24](https://github.com/prashar32/riskkernel/issues/24))
-  and local Ollama ([#23](https://github.com/prashar32/riskkernel/issues/23)).
-- **Streaming proxy** — SSE pass-through with mid-stream budget enforcement
-  ([#22](https://github.com/prashar32/riskkernel/issues/22)).
-- **OTLP ingress** — consume GenAI spans to govern apps RiskKernel didn't instrument
-  ([#90](https://github.com/prashar32/riskkernel/issues/90)).
-- **Postgres storage backend** — behind the same `Store` interface, SQLite stays the
-  default ([#25](https://github.com/prashar32/riskkernel/issues/25)).
-- **Operability** — a `/metrics` endpoint ([#91](https://github.com/prashar32/riskkernel/issues/91)),
-  a `riskkernel doctor` setup check ([#94](https://github.com/prashar32/riskkernel/issues/94)),
-  shell completions ([#93](https://github.com/prashar32/riskkernel/issues/93)),
-  and a Homebrew tap ([#97](https://github.com/prashar32/riskkernel/issues/97)).
+- **More native providers** — AWS Bedrock ([#24](https://github.com/prashar32/riskkernel/issues/24));
+  the long tail via LiteLLM-as-upstream.
+- **More backend dashboards** — a Datadog dashboard to join the Grafana and SigNoz
+  examples.
+- **Easier install** — a Homebrew tap for `brew install riskkernel`
+  ([#97](https://github.com/prashar32/riskkernel/issues/97)).
 
 ## Exploring