prisis · Oct 6, 2026 · May 9, 2026 · May 6, 2026 · Jul 20, 2026 · Jul 20, 2026
diff --git a/README.md b/README.md
@@ -93,6 +93,11 @@ your key, never the key itself. See
 [project link and credentials](https://github.com/vercel-labs/deepsec/blob/main/docs/vercel-setup.md)
 for the full reference.
 
+When running locally, `deepsec` can also reuse existing Claude, Codex, Pi, or
+OpenCode provider authentication — including a local `opencode` CLI login
+(`opencode` → `/connect`) — for evaluation-scale scans.
+
+
 If a `process` or `revalidate` run halts because the upstream credential
 ran out of quota or credits, deepsec stops gracefully and tells you
 where to top up. Re-run the same command afterward and it picks up

diff --git a/docs/architecture.md b/docs/architecture.md
@@ -105,12 +105,13 @@ Generated specs must use unique slugs; collisions are rejected before scan.
 - **Outputs:** FileRecord `findings[]` populated, `status: "analyzed"`,
   `analysisHistory[]` appended.
 
-Three agent backends are supported:
+Four agent backends are supported:
 
 | `--agent` | SDK | Default model |
 |---|---|---|
 | `codex` (default) | `@openai/codex-sdk` | `gpt-5.5` |
 | `claude` | `@anthropic-ai/claude-agent-sdk` | `claude-opus-4-8` |
+| `opencode` | `@opencode-ai/sdk/v2` | `anthropic/claude-opus-4-8` |
 | `pi` | `@earendil-works/pi-coding-agent` | `zai/glm-5.2` |
 
 Same prompt, same JSON output schema. You can mix backends within a

diff --git a/docs/configuration.md b/docs/configuration.md
@@ -31,7 +31,7 @@ see [`samples/webapp/deepsec.config.ts`](https://github.com/vercel-labs/deepsec/
 | `projects` | `ProjectDeclaration[]` | The codebases deepsec knows about. |
 | `plugins` | `DeepsecPlugin[]` | Loaded in order; later plugins override single-slot capabilities. |
 | `matchers` | `{ only?: string[]; exclude?: string[] }` | Filter the matcher set used by `scan`. |
-| `defaultAgent` | `string` | Default `--agent` value (`codex`, `claude`, or `pi`). See [models](models.md). |
+| `defaultAgent` | `string` | Default `--agent` value (`codex`, `claude`, `opencode`, or `pi`). See [models](models.md). |
 | `defaultModel` | `string` | Default `--model` value selected during setup. |
 | `defaultThinkingLevel` | `string` | Default reasoning effort (`minimal` through `xhigh`) selected during setup. |
 | `ai` | `ModelRoute` | Non-secret model credential route selected and verified by setup. |
@@ -164,19 +164,21 @@ requires the complete access-token triple.
 
 | Var | Used by | Purpose |
 |---|---|---|
-| `AI_GATEWAY_API_KEY` | Gateway route | Optional long-lived alternative to linked-project OIDC. Expanded for the selected agent. |
-| `ANTHROPIC_AUTH_TOKEN` | `process`, `revalidate`, `triage` (Claude backend) | API token for the Claude Agent SDK. AI Gateway-issued or Anthropic-issued. Set this if you don't use `AI_GATEWAY_API_KEY`. |
+| `AI_GATEWAY_API_KEY` | Gateway route | Optional long-lived alternative to linked-project OIDC. Expanded for the selected agent (Claude, Codex, OpenCode, or Pi). |
+| `ANTHROPIC_AUTH_TOKEN` | `process`, `revalidate`, `triage` (Claude or OpenCode Anthropic backend) | API token for the Claude Agent SDK or the OpenCode Anthropic provider. AI Gateway-issued or Anthropic-issued. Set this if you don't use `AI_GATEWAY_API_KEY`. |
+| `ANTHROPIC_API_KEY` | same | Direct-Anthropic credential (`x-api-key` broker contract). Prefer this over `ANTHROPIC_AUTH_TOKEN` when pointing directly at `https://api.anthropic.com` — some tooling rejects bearer-style tokens from that endpoint. |
 | `ANTHROPIC_BASE_URL` | same | Default (when `AI_GATEWAY_API_KEY` is set): `https://ai-gateway.vercel.sh`. Set to `https://api.anthropic.com` for direct Anthropic. |
 | `<ai.apiKeyEnv>` | Direct/custom route | User-chosen variable containing the provider credential. The name is stored in config; the value comes from `.env.local` or the process. |
 
 ### Optional
 
 | Var | Used by | Purpose |
 |---|---|---|
-| `OPENAI_API_KEY` | `--agent codex`, `--agent pi --model openai/...` | Codex SDK token or Pi OpenAI-provider token. Unset is fine if `AI_GATEWAY_API_KEY` is set. |
-| `OPENAI_BASE_URL` | `--agent codex` | Default (when `AI_GATEWAY_API_KEY` is set): `https://ai-gateway.vercel.sh/v1`. |
+| `OPENAI_API_KEY` | `--agent codex`, `--agent opencode --model openai/...`, `--agent pi --model openai/...` | Codex SDK, OpenCode, or Pi OpenAI-provider token. Unset is fine if `AI_GATEWAY_API_KEY` is set. |
+| `OPENAI_BASE_URL` | `--agent codex`, `--agent opencode --model openai/...` | Default (when `AI_GATEWAY_API_KEY` is set): `https://ai-gateway.vercel.sh/v1`. |
 | `PI_CODING_AGENT_DIR` | `--agent pi` | Optional Pi config/auth directory. Defaults to `~/.pi/agent`; local non-sandbox runs can reuse `auth.json` there. |
-| `DEEPSEC_AGENT_DEBUG` | both backends | Set to `1` to enable verbose agent logging. |
+| `XDG_DATA_HOME` | `--agent opencode` | Optional OpenCode data root. Local preflight looks for provider auth at `$XDG_DATA_HOME/opencode/auth.json` (default `~/.local/share/opencode/auth.json`). |
+| `DEEPSEC_AGENT_DEBUG` | AI backends | Set to `1` to enable verbose agent logging. |
 | `DEEPSEC_DATA_ROOT` | core | Override the data directory location. Equivalent to `dataDir` in config. |
 
 ### Plugin-specific

diff --git a/docs/data-layout.md b/docs/data-layout.md
@@ -165,7 +165,7 @@ deleted.
 | `investigatedAt` | `string` (ISO) | Timestamp. |
 | `durationMs` | `number` | Wall-clock total. |
 | `durationApiMs` | `number?` | API time only (excludes process orchestration). |
-| `agentType` | `string` | `claude-agent-sdk` or `codex`. |
+| `agentType` | `string` | Producing backend: `claude-agent-sdk`, `codex`, `opencode`, `pi`, or a plugin agent type. |
 | `model` | `string` | Model identifier. |
 | `modelConfig` | `Record<string, unknown>` | Provider-specific settings echoed back. |
 | `agentSessionId` | `string?` | The agent's session/thread id, for reproducing or replaying. |

diff --git a/docs/models.md b/docs/models.md
@@ -1,16 +1,17 @@
 ---
 title: "Models"
-description: "Choose Codex, Claude, or Pi for process and revalidate runs, and compare models under the same workload."
+description: "Choose Codex, Claude, OpenCode, or Pi for process and revalidate runs, and compare models under the same workload."
 ---
 
 deepsec talks to LLMs through interchangeable agent backends:
 
-| Backend                     | Default model         | Used by                      |
-|-----------------------------|-----------------------|------------------------------|
-| `codex` (default)           | `gpt-5.5`             | `process`, `revalidate`      |
-| `claude`                    | `claude-opus-4-8`     | `process`, `revalidate`      |
-| `pi`                        | `zai/glm-5.2`        | `process`, `revalidate` |
-| `claude` (triage)           | `claude-sonnet-4-6`   | `triage` (Claude-only)       |
+| Backend                     | Default model                   | Used by                 |
+|-----------------------------|---------------------------------|-------------------------|
+| `codex` (default)           | `gpt-5.5`                       | `process`, `revalidate` |
+| `claude`                    | `claude-opus-4-8`               | `process`, `revalidate` |
+| `opencode`                  | `anthropic/claude-opus-4-8`     | `process`, `revalidate` |
+| `pi`                        | `zai/glm-5.2`                   | `process`, `revalidate` |
+| `claude` (triage)           | `claude-sonnet-4-6`             | `triage` (Claude-only)  |
 
 Interactive one-shot setup recommends five benchmark-backed combinations:
 GPT-5.6 Sol, Claude Opus 5, Kimi K3, Grok 4.5, and the current DeepSeek entry.
@@ -55,6 +56,12 @@ disables the env-var preflight checks — the machine-wide login is used
 directly. Sandbox commands are the exception: they must broker a real token,
 so they still require `AI_GATEWAY_API_KEY` (or an equivalent key).
 
+
+The built-in backends also work with direct `AI_GATEWAY_API_KEY` credentials:
+one key covers Codex, Claude, OpenCode, and Pi. OpenCode and Pi accept
+provider/model identifiers directly, which makes them useful for comparing
+harness and provider behavior under the same deepsec workload.
+
 ## CLI selection
 
 ```bash
@@ -70,6 +77,10 @@ pnpm deepsec process --project-id my-app --agent codex
 # Codex backend, specific model:
 pnpm deepsec process --project-id my-app --agent codex --model gpt-5.4
 
+# OpenCode SDK harness (provider/model is required):
+pnpm deepsec process --project-id my-app --agent opencode
+pnpm deepsec process --project-id my-app --agent opencode --model openai/gpt-5.5
+
 # Pi backend through Vercel AI Gateway, default model:
 pnpm deepsec process --project-id my-app --agent pi
 
@@ -101,11 +112,12 @@ over large repos.
 
 The flag maps onto each backend's native dial:
 
-| Backend  | Setting                                     |
-|----------|---------------------------------------------|
-| `codex`  | model reasoning effort (`minimal`–`xhigh`)  |
-| `pi`     | thinking level (`minimal`–`xhigh`)          |
-| `claude` | adaptive-thinking effort (`minimal` → `low`, `xhigh` → `max`) |
+| Backend    | Setting                                                       |
+|------------|---------------------------------------------------------------|
+| `codex`    | model reasoning effort (`minimal`–`xhigh`)                    |
+| `opencode` | provider variant (Anthropic maps to `high` / `max`)           |
+| `pi`       | thinking level (`minimal`–`xhigh`)                            |
+| `claude`   | adaptive-thinking effort (`minimal` → `low`, `xhigh` → `max`) |
 
 It applies to the main investigation/revalidation runs only.
 Special-purpose follow-up calls (the refusal report, JSON repair) keep
@@ -171,6 +183,62 @@ Later `process`, `revalidate`, and Sandbox commands resolve the persisted
 route. Per-command `--ai-provider`, `--ai-base-url`, `--ai-api-key-env`, and
 repeatable `--ai-header name=value` remain available as Pi runtime overrides.
 
+### OpenCode SDK harness
+
+OpenCode uses `@opencode-ai/sdk/v2` plus the `opencode` runtime — both the
+v1 runtime (npm `opencode-ai` 1.x, bundled for sandbox workers) and the
+**OpenCode v2** runtime (2.0.x) are supported; the generation is detected at
+startup and the matching protocol is used automatically. Each batch starts a
+private, password-secured local OpenCode server, creates a session rooted at
+the target project, and asks for JSON output. The deepsec agent allows only
+`read`, `glob`, `grep`, and `list`; shell, edits, network tools, subagents,
+external directories, LSP, skills, and MCP-triggered permissions are denied.
+The session and server are closed on success, error, or abort.
+
+Differences worth knowing:
+
+- The v1 runtime returns JSON Schema–validated structured output when the
+  model supports it; the v2 runtime has no structured-output prompts, so
+  results arrive as plain text and go through deepsec's shared JSON-repair
+  pipeline (same contract, same validation).
+- The v2 runtime always requires basic auth on its server; deepsec generates
+  the credential itself and nothing is written to disk.
+- `--ai-header` overrides require the v1 runtime; on v2 they are rejected
+  with a clear error.
+
+**Vercel AI Gateway on the v2 runtime** — two supported options:
+
+- Native gateway provider (recommended): the v2 model catalog ships a
+  `vercel` provider ("Vercel AI Gateway") that reads `AI_GATEWAY_API_KEY`
+  directly — one key covers every model behind the gateway:
+
+  ```bash
+  AI_GATEWAY_API_KEY=vck_… pnpm deepsec process --project-id my-app \
+    --agent opencode \
+    --model vercel/anthropic/claude-opus-4-8
+  ```
+
+- The standard `ANTHROPIC_*`/`OPENAI_*` expansion also keeps working:
+  deepsec translates the expansion into the v2 provider `baseURL` overlays
+  and bridges the gateway bearer token to the `x-api-key` form the v2
+  anthropic provider reads, so `--model anthropic/claude-opus-4-8` with
+  `AI_GATEWAY_API_KEY` routes through the gateway unchanged.
+
+Models use OpenCode's required `provider/model` form:
+
+```bash
+pnpm deepsec process --project-id my-app \
+  --agent opencode \
+  --model anthropic/claude-opus-4-8
+```
+
+For a local run without environment credentials, authenticate a provider in
+OpenCode first: run `opencode`, then use `/connect`. Sandbox runs cannot reuse
+that local credential store; use AI Gateway or an explicit provider token.
+Generic `--ai-provider`, `--ai-base-url`, `--ai-api-key-env`, and
+`--ai-header` overrides work for OpenCode as they do for Pi (headers v1
+runtime only).
+
 ### `claude-sonnet-4-6` for `triage`
 
 Triage buckets findings into P0/P1/P2/skip without re-reading the code.
@@ -216,6 +284,7 @@ rest of deepsec stays unchanged:
 ```bash
 pnpm deepsec process --project-id my-app --model anthropic-mythos-1
 pnpm deepsec process --project-id my-app --agent codex --model gpt-6
+pnpm deepsec process --project-id my-app --agent opencode --model anthropic/claude-mythos-1
 pnpm deepsec process --project-id my-app --agent pi --model vercel-ai-gateway/openai/gpt-6
 ```
 

diff --git a/docs/vercel-setup.md b/docs/vercel-setup.md
@@ -100,7 +100,7 @@ ai: { mode: "gateway", provider: "vercel" }
 ```
 
 Deepsec maps the Gateway credential to the environment expected by Codex,
-Claude, or Pi. Sandbox workers receive only a placeholder. The real bearer
+Claude, OpenCode, or Pi. One key covers all four harnesses. Sandbox workers receive only a placeholder. The real bearer
 token is injected at the allowed Gateway host by the host-side broker.
 
 ### Your own OpenAI or Anthropic credential
@@ -133,13 +133,14 @@ again or put it in `.deepsec/.env.local`:
 MY_OPENAI_KEY=...
 ```
 
-Direct OpenAI routes require Codex; direct Anthropic routes require Claude.
+Direct OpenAI routes require Codex or OpenCode (`openai/...` models); direct
+Anthropic routes require Claude or OpenCode (`anthropic/...` models).
 Setup rejects incompatible agent/provider combinations before scanning.
 
 ### Custom HTTPS provider
 
-Custom routes are supported by the Pi backend. Supply an HTTPS base URL and
-describe how the credential should be attached:
+Custom routes are supported by the Pi and OpenCode backends. Supply an HTTPS
+base URL and describe how the credential should be attached:
 
 ```bash
 MARTIAN_KEY=... npx deepsec init \
@@ -156,6 +157,29 @@ Use `:bearer` for `Authorization: Bearer …` and `:raw` for provider-specific
 raw token headers. HTTP URLs, embedded URL credentials, invalid header names,
 and custom routes for Codex/Claude are rejected.
 
+OpenCode uses `provider/model` identifiers, so the same custom route works
+there too:
+
+```bash
+MARTIAN_KEY=... npx deepsec init \
+  --agent opencode \
+  --model openai/gpt-5.5 \
+  --model-auth custom \
+  --ai-provider martian \
+  --ai-api-key-env MARTIAN_KEY \
+  --ai-base-url https://api.martian.example/v1 \
+  --ai-credential-header authorization:bearer
+```
+
+### Local agent authentication
+
+Non-sandbox runs can reuse an existing local login for the built-in
+harnesses: `claude login` / `codex login` for those CLIs, or run `opencode`
+and use `/connect` to connect a provider for the OpenCode backend. Local
+OpenCode auth (stored under `$XDG_DATA_HOME/opencode/auth.json`) is never
+copied into Vercel Sandbox workers — sandbox runs need the Gateway (or a
+direct/custom route).
+
 ## Credential brokering in Sandbox
 
 Deepsec resolves the real model credential on the host. A worker receives a
@@ -211,5 +235,6 @@ pnpm deepsec sandbox-all process --sandboxes 30 --concurrency 4
 | Direct route asks for a custom variable | Config stores its name, not its value. | Export it in the fresh shell or add it to `.env.local`. |
 | Setup repeats login verification | The link, route, agent, or verification age changed. | Complete the probe; later unchanged runs short-circuit again. |
 | Gateway quota is exhausted | Processing stopped before launching more batches. | Add credits or change route, then re-run; processing resumes pending files. |
+| `Missing AI credentials for --agent opencode` | No Gateway/direct credential and no local OpenCode auth. | Set `AI_GATEWAY_API_KEY`, use a direct route, or run `opencode` and use `/connect` (local runs only). |
 
 References: [AI Gateway authentication](https://vercel.com/docs/ai-gateway/authentication-and-byok#quick-start), [Sandbox authentication](https://vercel.com/docs/vercel-sandbox/concepts/authentication).
diff --git a/e2e/bundle.test.ts b/e2e/bundle.test.ts
@@ -47,6 +47,10 @@ describe("bundle e2e", () => {
     expect(stdout).toContain("deepsec");
     expect(stdout).toContain("scan");
     expect(stdout).toContain("process");
+
+    const processHelp = runBundle(["process", "--help"]);
+    expect(processHelp.status).toBe(0);
+    expect(processHelp.stdout).toContain("opencode");
   });
 
   it("config.d.ts is self-contained (no internal @deepsec/* re-exports)", () => {
@@ -186,7 +190,9 @@ export default defineConfig({
       expect(stdout).toContain("webapp-debug-flag");
       expect(stdout).toContain("webapp-route-no-rate-limit");
     } finally {
-      fs.rmSync(link, { force: true });
+      try {
+        fs.unlinkSync(link);
+      } catch {}
       fs.rmSync(path.join(sampleDir, "data"), { recursive: true, force: true });
     }
   });

diff --git a/e2e/pipeline-sandbox.test.ts b/e2e/pipeline-sandbox.test.ts
@@ -64,6 +64,7 @@ const HAS_SANDBOX_KEY =
     Boolean(process.env.VERCEL_PROJECT_ID));
 
 const SHOULD_RUN = LIVE && HAS_SANDBOX_KEY;
+const SANDBOX_TIMEOUT_MS = 30 * 60_000;
 
 /**
  * Stub agent definition shared with pipeline.test.ts. Inlined here
@@ -319,6 +320,8 @@ describe.skipIf(!SHOULD_RUN)("pipeline e2e — live sandbox", () => {
             "1",
             "--vcpus",
             "2",
+            "--timeout",
+            String(SANDBOX_TIMEOUT_MS),
             "--limit",
             "2",
             "--concurrency",
@@ -358,6 +361,8 @@ describe.skipIf(!SHOULD_RUN)("pipeline e2e — live sandbox", () => {
             "1",
             "--vcpus",
             "2",
+            "--timeout",
+            String(SANDBOX_TIMEOUT_MS),
             "--limit",
             "2",
             "--concurrency",

diff --git a/knip.json b/knip.json
@@ -35,6 +35,9 @@
     "@anthropic-ai/claude-agent-sdk",
     "@earendil-works/pi-coding-agent",
     "@openai/codex",
-    "@openai/codex-sdk"
+    "@openai/codex-sdk",
+    "@opencode-ai/sdk",
+    "opencode-ai",
+    "undici"
   ]
 }