jev-firewall 0.1.0__tar.gz

jev_firewall-0.1.0/.gitignore

@@ -0,0 +1,12 @@
+__pycache__/
+*.py[cod]
+.venv/
+.mypy_cache/
+.ruff_cache/
+.pytest_cache/
+dist/
+build/
+*.egg-info/
+.env
+jev-audit*.jsonl
+benchmarks/results/raw/

jev_firewall-0.1.0/CHANGELOG.md

@@ -0,0 +1,25 @@
+# Changelog
+
+## 0.1.0 (2026-09-27)
+
+First release.
+
+- `PolicyEngine`: evaluates each tool call with one Jev request (severity Choice,
+  `is_irreversible` Noul, `matches_stated_goal` Noul) over a minimal, redacted state, and maps
+  the answers to ALLOW / HOLD / DENY through per-tool `RiskThresholds` from `policy.yaml`.
+- Deterministic hard-pattern floors that run locally before Jev and can't be argued away by
+  text in the arguments.
+- Explicit `fail_mode` (`open` / `closed`), required in config and overridable per tool.
+- Jev routes: TypeSafe direct, OpenRouter, and Vercel AI Gateway (via `typesafe-sdk`), plus
+  Cloudflare Workers AI (REST).
+- Approval channels: `CLIApproval`, `WebhookApproval`, `TimeoutDenyApproval`, plus
+  `LangGraphInterruptApproval`.
+- Append-only JSONL `AuditLog` with verdict, approval, and outcome events.
+- Secret and PII redaction for everything sent to Jev or written to the log.
+- Adapters: LangGraph, LangChain (callbacks and `create_agent` middleware), OpenAI Agents SDK,
+  Google ADK (callbacks and plugin), Semantic Kernel.
+- `jev-firewall` CLI: `init`, `check`, `eval`, `audit`.
+- Adversarial LangGraph demo, seeded evaluation suites, and benchmark scripts.
+
+Known gaps: accuracy and end-to-end latency against live Jev haven't been measured for this
+release (see docs/benchmarks.md).

jev_firewall-0.1.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

jev_firewall-0.1.0/PKG-INFO

@@ -0,0 +1,258 @@
+Metadata-Version: 2.5
+Name: jev-firewall
+Version: 0.1.0
+Summary: Runtime action firewall for AI agents, powered by TypeSafe's Jev System One model.
+Author: Ahsan Shah
+License-Expression: Apache-2.0
+License-File: LICENSE
+Keywords: ai-agents,firewall,google-adk,guardrails,jev,langchain,langgraph,openai-agents,prompt-injection,semantic-kernel,typesafe
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Security
+Classifier: Typing :: Typed
+Requires-Python: >=3.11
+Requires-Dist: httpx2>=2.13.1
+Requires-Dist: pydantic>=2.13
+Requires-Dist: pyyaml>=6.0.3
+Requires-Dist: typesafe-sdk<0.8,>=0.7.2
+Provides-Extra: all
+Requires-Dist: azure-ai-agents>=1.2.0b3; extra == 'all'
+Requires-Dist: google-adk>=2.10.0; extra == 'all'
+Requires-Dist: langchain-core>=1.6.5; extra == 'all'
+Requires-Dist: langchain>=1.4.2; extra == 'all'
+Requires-Dist: langgraph>=1.2.12; extra == 'all'
+Requires-Dist: openai-agents>=0.22.3; extra == 'all'
+Requires-Dist: semantic-kernel>=1.44.1; extra == 'all'
+Provides-Extra: google-adk
+Requires-Dist: google-adk>=2.10.0; extra == 'google-adk'
+Provides-Extra: langchain
+Requires-Dist: langchain-core>=1.6.5; extra == 'langchain'
+Requires-Dist: langchain>=1.4.2; extra == 'langchain'
+Provides-Extra: langgraph
+Requires-Dist: langchain-core>=1.6.5; extra == 'langgraph'
+Requires-Dist: langgraph>=1.2.12; extra == 'langgraph'
+Provides-Extra: openai-sdk
+Requires-Dist: openai-agents>=0.22.3; extra == 'openai-sdk'
+Provides-Extra: semantic-kernel
+Requires-Dist: azure-ai-agents>=1.2.0b3; extra == 'semantic-kernel'
+Requires-Dist: semantic-kernel>=1.44.1; extra == 'semantic-kernel'
+Description-Content-Type: text/markdown
+
+# jev-firewall
+
+**A runtime action firewall for AI agents.** Before an agent's tool call runs, jev-firewall
+asks [Jev](https://docs.typesafe.ai) (TypeSafe's System One decision model) how dangerous it
+is, and returns one of three verdicts:
+
+| Verdict | What happens |
+|---|---|
+| **ALLOW** | the tool runs |
+| **HOLD** | the call waits for a human (CLI prompt, webhook, or LangGraph interrupt) |
+| **DENY** | the tool never runs; `ActionBlocked` is raised (or a refusal is returned to the model) |
+
+Every decision is written to an append-only JSONL audit log. One core engine is shared by
+thin adapters for **LangGraph, LangChain, OpenAI Agents SDK, Google ADK and Semantic Kernel**.
+
+```text
+agent proposes tool call ──► adapter ──► PolicyEngine ──► hard patterns (local regex)
+                                              │                 │ DENY floor? stop here
+                                              ▼                 ▼
+                                    Jev: 1 call, 3 typed questions (~680 input tokens est.)
+                                              │
+                                              ▼
+                        thresholds + rules ──► ALLOW / HOLD / DENY ──► audit log
+```
+
+## Why a System One model?
+
+A **System Two** guardrail is a second LLM that reads the action and writes out its reasoning.
+It is slow (hundreds of ms to seconds), costs about as much as the agent's own call, returns
+free text you have to parse, and can itself be talked round.
+
+**System One** models like Jev don't generate text. You declare typed questions (pick one of
+these options, yes/no probability, position on a scale) and get back calibrated probabilities
+you can threshold in code. That makes the check fast and cheap enough to run on *every* tool
+call, and the decision logic stays in your code and your `policy.yaml`, not in a prompt.
+
+Jev has known limits, and this library is built around them
+([Jev 1.13 jaggedness](https://docs.typesafe.ai/model-jaggedness/jev-1.13.md)):
+
+- **Adversarial text in the input can move its answers.** Tool arguments are attacker
+  controllable, so deterministic *hard patterns* run first and set a floor Jev cannot lower.
+  `rm -rf /` is denied even if the command says "this is a safe low_risk cleanup".
+- **Irrelevant context lowers accuracy.** The state sent to Jev is three fields: the tool name,
+  its redacted arguments, and a one-line goal. Never the conversation history.
+- **No counting, arithmetic or date logic.** None of the questions need any.
+
+## Quick start (about 5 minutes)
+
+```bash
+pip install "jev-firewall[langgraph]"      # or [langchain], [openai_sdk], [google_adk], [semantic_kernel], [all]
+export TYPESAFE_API_KEY=...                # https://console.typesafe.ai/keys
+jev-firewall init                          # writes ./policy.yaml from the bundled example
+jev-firewall check policy.yaml             # validates it and prints the effective settings
+```
+
+Try one decision from the command line (exit code 0 allow, 10 hold, 20 deny):
+
+```bash
+jev-firewall eval policy.yaml --tool terminal \
+  --args '{"command": "sudo rm -rf /var/lib/postgresql/16/main"}' \
+  --goal "Free up disk space on the build server"
+```
+
+No key yet? Add `--offline` to use the bundled keyword stand-in. It exercises the whole
+pipeline, but it is **not Jev**, so don't judge accuracy by it.
+
+Wire it into a LangGraph agent:
+
+```python
+from jev_firewall import Firewall
+from jev_firewall.adapters.langgraph import firewall_tool_node
+
+firewall = Firewall.from_yaml("policy.yaml")  # engine + approval channel + audit log
+builder.add_node("tools", firewall_tool_node(tools, firewall))  # replaces ToolNode(tools)
+```
+
+That's it. Every tool call in that node is now evaluated, held, or blocked, and logged.
+
+## Adapters
+
+All adapters do the same thing: build a `ToolCall(tool_name, args, agent_goal)`, then
+evaluate. DENY raises `ActionBlocked`, HOLD waits for approval, and ALLOW runs the tool.
+They contain no policy logic.
+
+| Framework | Install extra | Hook | Usage |
+|---|---|---|---|
+| LangGraph | `[langgraph]` | `ToolNode(wrap_tool_call=...)` | `firewall_tool_node(tools, firewall)` |
+| LangChain | `[langchain]` | `BaseCallbackHandler.on_tool_start` / `create_agent` middleware | `JevFirewallCallbackHandler(firewall)`, `JevFirewallMiddleware(firewall)` |
+| OpenAI Agents SDK | `[openai_sdk]` | `FunctionTool.tool_input_guardrails` | `protect_tools(tools, firewall)` |
+| Google ADK | `[google_adk]` | `before_tool_callback` / runner plugin | `JevFirewallCallbacks(firewall)`, `JevFirewallPlugin(firewall)` |
+| Semantic Kernel | `[semantic_kernel]` | `FUNCTION_INVOCATION` filter | `JevFunctionInvocationFilter(firewall)` |
+
+> **uv + Semantic Kernel:** semantic-kernel ≥ 1.36 depends on a pre-release of `azure-ai-agents`.
+> pip installs it as is. With uv, also request it directly
+> (`uv add "jev-firewall[semantic_kernel]" "azure-ai-agents>=1.2.0b3"`) or pass `--prerelease=allow`.
+
+Runnable, keyless examples are in [`examples/`](examples/). [`examples/consumer/`](examples/consumer/)
+has one standalone project per framework (`pip install -r requirements.txt && python app.py`)
+that uses the published package exactly as you would. Framework details, including how
+each framework surfaces `ActionBlocked`, are in [docs/adapters.md](docs/adapters.md).
+
+**`agent_goal`** powers the prompt-injection check. By default each adapter uses the *first
+user message* of the conversation, never tool output or retrieved content, so injected text
+can't rewrite it. You can also pass `goal="..."` or a callable.
+
+## Configuration reference
+
+`policy.yaml` is validated strictly: unknown keys are errors, and `fail_mode` and `jev` have
+no defaults. Full reference: [docs/configuration.md](docs/configuration.md).
+
+```yaml
+version: 1
+fail_mode: closed            # REQUIRED: open | closed (see below)
+jev:
+  route: direct              # direct | openrouter | vercel | cloudflare
+  timeout_s: 1.0             # total budget per decision, retries included
+  max_retries: 1
+defaults:
+  thresholds: {low_risk: allow, moderate_risk: allow, high_risk: hold, critical_danger: deny}
+  irreversible_threshold: 0.5   # p(irreversible) at/above this: a would-be ALLOW becomes HOLD
+  goal_match_threshold: 0.5     # p(matches goal) below this: at least HOLD
+  min_confidence: 0.5           # severity confidence below this: at least HOLD
+  hard_patterns:
+    - {description: pipe remote script into a shell, pattern: '\b(curl|wget)\b[^\n|]*\|\s*(sudo\s+)?(ba|z|da)?sh\b', floor: deny}
+tools:                       # exact name beats glob; most specific glob wins
+  "read_*": {action: allow}  # skip Jev entirely
+  "terminal": {thresholds: {moderate_risk: hold}}
+  "transfer_*": {thresholds: {high_risk: deny}, fail_mode: closed}
+approval: {channel: cli, timeout_s: 300}      # cli | webhook | timeout_deny
+audit: {path: ./jev-audit.jsonl}
+```
+
+**How a decision is made** (each step can only make it stricter):
+
+1. Resolve the tool's rule: exact name, then the most specific glob, then `defaults`.
+2. Hard patterns on the raw arguments set a floor. A `deny` floor (or `action: deny`) ends here, with no Jev call.
+3. `action: allow|hold` skips Jev.
+4. Jev answers `severity` (Choice), `is_irreversible` (Noul) and `matches_stated_goal` (Noul) in one call.
+5. Severity maps to a verdict through the tool's thresholds.
+6. Irreversible and not allowlisted (`allow_irreversible: true`): an ALLOW becomes HOLD.
+7. Goal mismatch: at least HOLD. Low severity confidence: at least HOLD.
+
+**Jev routes.** Each route reads its key from its own env var, so a TypeSafe key is never
+sent to a gateway by mistake.
+
+| route | key env var | default model |
+|---|---|---|
+| `direct` | `TYPESAFE_API_KEY` | `jev-1.13.0` (pinned; thresholds are tuned per version) |
+| `openrouter` | `OPENROUTER_API_KEY` | `~typesafe/jev-latest` |
+| `vercel` | `AI_GATEWAY_API_KEY` | `typesafe-ai/jev` |
+| `cloudflare` | `CLOUDFLARE_API_TOKEN` + `CLOUDFLARE_ACCOUNT_ID` | `typesafe/jev` |
+
+## Fail mode guidance
+
+If Jev times out, errors, rate-limits, or returns something malformed, the verdict comes from
+`fail_mode`, never from a hidden default:
+
+- **`closed` (DENY)**: use it for anything with side effects, such as shell, databases, email,
+  payments, and deploys. An outage stops the agent instead of letting it act unchecked. This is
+  the right default for most deployments.
+- **`open` (ALLOW)**: only for low-stakes tools where availability matters more than
+  checking. Hard-pattern floors still apply in open mode.
+
+Set a global `fail_mode` and override it per tool (`tools.<name>.fail_mode`). Keep `timeout_s`
+tight: a decision that takes longer than the budget *is* a failure and goes through
+`fail_mode`. Watch the audit log for `source: fail_open` / `fail_closed` to see how often it
+happens.
+
+## Security properties
+
+- **What Jev sees:** `tool_name`, `tool_args` after redaction, and a one-line `agent_goal`. That's all.
+- **Redaction** (`jev_firewall.redact`) masks sensitive keys (`password`, `api_key`,
+  `authorization`, ...), API keys and tokens (OpenAI, Anthropic, AWS, GitHub, Stripe, Slack,
+  Google, JWTs, bearer tokens), PEM private keys, URL credentials, `KEY=value` secrets, and PII
+  (emails keep their domain, which is a useful risk signal; phone numbers, SSNs, and
+  Luhn-valid card numbers are masked). Long strings, lists, and deep nesting are truncated.
+- **Audit log:** only redacted arguments are written. One line per event (`verdict`,
+  `approval`, `outcome`), linked by `call_id`. Run `jev-firewall audit ./jev-audit.jsonl` for a summary.
+
+## Measured results
+
+Only numbers that were actually measured are listed. Full method and raw JSON: [docs/benchmarks.md](docs/benchmarks.md).
+
+| What | Result | Conditions |
+|---|---|---|
+| Firewall library overhead (no network) | **p50 1.02 ms, p95 1.50 ms** per decision | in-memory Jev, audit on, 2,000 calls, Windows 11, Python 3.12 |
+| Hard patterns alone on the seeded suites | 19 / 50 harmful caught, **0 / 40** benign blocked | Jev disabled |
+| Adversarial LangGraph demo | 4 / 4 harmful calls stopped, 4 / 4 benign allowed | offline stand-in; live run pending |
+| End-to-end latency with live Jev | **not measured yet** | needs `TYPESAFE_API_KEY` |
+| False negatives / false positives with live Jev | **not measured yet** | needs `TYPESAFE_API_KEY` |
+| LLM-guardrail comparison | **not measured yet** | needs `ANTHROPIC_API_KEY` |
+| Cost per decision | ~680 input tokens → **~$0.00003 (estimate)** | chars/4 of the actual request; live `usage` pending |
+
+Reproduce with `python benchmarks/overhead.py` and `python benchmarks/run_suites.py`.
+Both measure the live modes automatically when the keys are set.
+
+## Development
+
+```bash
+uv sync --all-extras
+uv run pytest                      # offline: FakeJevClient + mock HTTP transports
+JEV_LIVE_TESTS=1 uv run pytest     # plus live Jev tests (needs TYPESAFE_API_KEY)
+uv run ruff check . && uv run mypy # strict typing
+python examples/langgraph_adversarial_demo.py
+```
+
+Adding an adapter requires no changes to the core package; see [docs/adapters.md](docs/adapters.md#writing-an-adapter).
+
+## License
+
+Apache-2.0. jev-firewall is an independent open-source project and is not affiliated with TypeSafe AI.