Wirken is the switchboard for the agent era. Message it on Telegram, Discord, Slack, Microsoft Teams, Matrix, WhatsApp, Signal, Google Chat, or iMessage, and the agent on the other end reads files, calls APIs, and runs tools on your behalf. Each channel gets its own line. Every call is logged before it connects.
Wirken ships as a single static Rust binary and works with Ollama, Anthropic, OpenAI, Gemini, Bedrock, Tinfoil, Privatemode, or any OpenAI-compatible endpoint. MIT licensed.
Each channel runs in its own adapter process with a distinct Ed25519 IPC identity and its own vault-scoped token set. Credentials sit in an XChaCha20-Poly1305 vault keyed from the OS keychain, with per-credential expiry and manual rotation tracked in the store. Every agent action, tool call, LLM request, and response is written to a per-session SHA-256 hash-chained audit log. The log forwards to Datadog, Splunk HEC, Microsoft Sentinel, or a webhook when SIEM is configured. Permissions follow a three-tier model scoped per agent. Parent agents that spawn children declare per-child ceilings: tool allowlist, maximum permission tier, max rounds, max runtime.
Install and run
Download the latest release binary:
curl -fsSL https://raw.githubusercontent.com/gebruder/wirken/main/install.sh | sh
wirken setup
wirken runPin the installer before piping. The committed install.sh has this SHA-256:
73e678196ea073608e902c8ab11a01ede07e0d37fddccaa20c43fa5d62bd52f5
Verify it yourself:
curl -fsSL https://raw.githubusercontent.com/gebruder/wirken/main/install.sh | sha256sumThe installer then fetches checksums.sha256 and checksums.sha256.sig from the release, verifies the signature with ssh-keygen -Y verify against a signing key embedded in the script, and verifies the binary's SHA-256 against the signed checksums. Every failure path is fail-closed: missing signature, missing checksum, mismatched digest, or a machine without sha256sum/shasum aborts install. The only override is WIRKEN_ALLOW_UNVERIFIED=1, which warns on stderr and is documented in docs/release-signing.md.
Prebuilt binaries are available for Linux (x86_64, aarch64), macOS (x86_64, Apple Silicon), and Windows 11 (x86_64). The Linux binaries are statically linked against musl with no glibc dependency. Windows users: the bash installer above does not apply; see docs/windows.md for the Windows install path and the feature-set differences (Signal adapter, orchestrator-push, service installer, and cron presets are Linux/macOS only).
Or build from source (requires Rust 1.85+ and the capnp compiler):
# Ubuntu/Debian sudo apt-get install -y capnproto # macOS brew install capnp # Windows choco install capnproto -y cargo install --path crates/cli
wirken setup walks you through three steps:
wirken setup
────────────
Step 1: Pick your AI
Provider: Ollama (local) / Anthropic / OpenAI / Google Gemini / AWS Bedrock / Tinfoil / Privatemode / Custom endpoint
API key: ********
Encrypting API key...
API key encrypted and stored.
Model: gpt-4.1-mini ← auto-detected from provider API
Step 2: Pick your channels
Add a channel: Telegram
Telegram bot token: ********
telegram: token encrypted, adapter keypair generated, registered.
Setup complete!
Provider: openai (gpt-4o)
Channels: telegram
wirken run starts the gateway daemon. It spawns adapter processes, accepts authenticated connections, routes messages to the agent, and serves a WebChat UI at http://localhost:18790:
wirken gateway v0.9.0
──────────────
Provider: ollama/llama3.2
Ollama version: 0.19.0
Route: telegram -> agent:default
Socket: ~/.wirken/sockets/gateway.sock
WebChat: http://localhost:18790
Gateway running. Press Ctrl+C to stop.
All local services bind to 127.0.0.1. Wirken never instructs you to bind inference servers, WebChat, or any local endpoint to 0.0.0.0.
Install as a system service so the gateway starts on login:
wirken setup --install-service
Architecture
graph TD
Channels["Telegram · Discord · Slack · Teams · Matrix · WhatsApp · Signal · Google Chat · iMessage"]
Channels -- "UDS · Ed25519 · Cap'n Proto" --> Registry
subgraph Gateway["Wirken"]
Registry[Adapter Registry] --> Router
Router --> Detect[Injection Detection]
Detect --> Factory[AgentFactory]
Factory --> Agent[Agent Runtime]
Agent --> Context[Context Engine]
Agent --> Permissions
Agent --> Skills
Agent --> Tools
Agent --> Vault --> Keychain
subgraph Execution
Tools --> Sandbox[Docker / gVisor / Wasm]
end
SessionLog["Session Log\n(per-session hash chain, attested)"]
end
Agent -- "UDS" --> McpProxy["MCP Proxy\n(separate process)"]
McpProxy -- "stdio · HTTP · OAuth2" --> McpServers[MCP Servers]
Agent -- HTTPS --> LLM[LLM Providers]
Agent -- "spawn_subagent" --> Factory
Detect -.-> SessionLog
Permissions -.-> SessionLog
Tools -.-> SessionLog
SessionLog -.-> SIEM[SIEM / Webhook]
Each channel adapter runs as a separate OS process. Adapters authenticate to the gateway with a per-adapter Ed25519 challenge-response handshake over a Unix domain socket. Messages are serialized with Cap'n Proto (zero-copy, traversal-limited). An adapter can only deliver inbound messages for its own channel and request outbound sends for its own channel. It cannot invoke tools, read other channels' sessions, or access other channels' credentials.
Channel isolation operates at two levels. The active mechanism is process-level: each channel runs in its own OS process with a distinct ed25519 identity. The IPC crate also defines a sealed Channel trait and SessionHandle<C: Channel> type that makes cross-channel handle conversions a compile error. This type-level API is not yet threaded through the production message path, where the channel discriminator is a string field on the Cap'n Proto inbound frame. If an adapter process is compromised, the blast radius is exactly one channel because the gateway's IPC boundary, running in a separate memory-safe process, prevents lateral movement.
The MCP proxy also runs out-of-process over a Unix domain socket, with the vault handle isolated in the proxy. MCP servers connect via stdio, HTTP, or OAuth2, and the agent process never sees MCP credentials.
Two distinct boundaries protect MCP credentials. Process isolation is per-MCP-server-config: every server entry in mcp.json spawns its own subprocess at the wirken UID, with its own stdin/stdout pipe and its own environment, and the agent process never holds plaintext bearer tokens or OAuth2 client secrets — those live behind the proxy's vault handle. Vault ACLs are per-credential-name: the credential store keys every entry by name (e.g. slack-token, github-pat), and a connector that asks for a credential it was not configured for receives a NotFound rather than a value from another connector's slot. The two are independent. A compromised MCP child cannot read another child's credentials because process isolation keeps every child to its own configured token set; a compromised credential name cannot be retrieved by an unrelated component because the vault store gates by name. Neither alone is sufficient, and the combination is what gives the MCP path its credential-scoping property.
Agents are stateless between turns. The AgentFactory wakes an agent for each inbound message by replaying its session log. Conversations are durably logged as typed session events (user messages, assistant messages, tool calls, tool results, LLM request/response metadata) in an append-only, per-session hash-chained table. If the agent crashes mid-turn, the harness detects incomplete tool rounds on wake and surfaces them as failures rather than silently re-executing side effects. A context engine trims conversations under each model's token budget before every LLM call, preferring to drop old tool results before touching user or assistant text.
Agents can delegate bounded subtasks to child agents via spawn_subagent. The operator configures a per-child capability ceiling (tool allowlist, max permission tier, max rounds, max runtime). Children run headless with no interactive approvals, isolated session logs, and a hard depth cap of 4.
Security properties
- Session attestation. Per-agent Ed25519 identity signs the per-session hash chain after every turn.
wirken session verifyreplays the log offline and re-checks message hashes, deterministic tool results, and chain integrity. Tampered sessions break the chain. - Reproducible replay. Every LLM call is recorded as a typed session event with a SHA-256 hash of the exact messages and tools sent. The verifier recomputes these hashes from the log and flags any divergence.
- Per-channel process isolation. Each channel adapter runs in its own OS process with a distinct ed25519 identity. Type-level channel separation (
SessionHandle<Telegram>vsSessionHandle<Discord>) exists in the IPC crate and is regression-tested but not yet used in the production message path. - Out-of-process credential isolation. MCP credentials (bearer tokens, OAuth2 client secrets) live in a separate proxy process. The agent process never sees them. The vault is XChaCha20-Poly1305, keyed from the OS keychain;
secrecy+zeroizemake logging a secret a compile error. - Capability-attenuated multi-agent. The LLM cannot widen a child agent's permissions. The operator sets the ceiling; the harness intersects, clamps, and enforces. Children run headless with no interactive approvals.
A handful of operator-controlled escape hatches relax specific
defaults: WIRKEN_ALLOW_UNSIGNED_ORG_CONFIG=1,
WIRKEN_ALLOW_UNSIGNED_SKILLS=1, sandbox.json mode off,
WIRKEN_WEBCHAT_ALLOW_NO_ORIGIN=1, and
WIRKEN_AUDIT_VERIFY_EVERY_FLUSHES=N (the last raises the
continuous-verify cadence; values above 10,000 effectively disable
the in-process verifier and emit a warn). Each is documented under
"Documented escape hatches" in the security-properties doc. Persistent flips need a
systemd EnvironmentFile, shell rc export, or wrapper script so the
operator's intent is durable across reboots.
Full OWASP and NIST AI RMF mappings: docs/security-properties.md
Enterprise deployment
Wirken gives organizations the controls they need to deploy AI agents without bypassing existing security, compliance, and audit requirements.
- Full attribution. Every inbound message records the platform sender id, channel, session, and agent. Permission decisions are scoped per agent, not per user. Typed session events record what action ran, when, and on which target.
- Tamper-evident audit trail. All actions logged as typed session events before execution. Per-session SHA-256 hash chain detects modification or deletion. Per-agent Ed25519 attestation signs the chain head after every turn.
wirken session verifyreplays the log offline and re-checks hashes. SIEM forwarding sends events to Datadog, Splunk HEC, Microsoft Sentinel, or any webhook in real time. - Crash recovery. Agents are stateless between turns. The harness replays the session log on wake. Incomplete tool rounds are detected and surfaced as failures rather than silently re-executed.
- Graduated permissions. Three-tier model. Workspace file access and web search are always allowed. Shell exec and external file access require first-use approval. Destructive operations, credential access, and skill installs always require explicit approval. Approvals expire after 30 days.
- Capability-attenuated multi-agent. Parent agents delegate to children via
spawn_subagentunder operator-configured ceilings (tool allowlist, max permission tier, max rounds, max runtime). Children run headless with isolated session logs. Hard depth cap of 4. - Sandboxed execution. Shell exec runs in an ephemeral Docker container by default (exec-only mode as of 0.7.5). Containers drop all Linux capabilities, set
no-new-privileges, use Docker's default seccomp profile, use a read-only root filesystem with a 64 MB tmpfs at/tmp, pin memory and PIDs, and run as a non-root user with no network. gVisor (runsc) is opt-in for kernel attack surface reduction. If Docker is not reachable at gateway start, the ToolRegistry logs a distinct warning and falls back to host execution for the agent's lifetime; if runsc is requested but not registered, the warning names that dependency specifically. Sandbox provisioning is lazy, so there is no startup cost when unused. Operators who setsandbox.jsontomode: offopt into runningexecat the wirken UID with no container; the gateway warns at startup, and the host shell can then read or rewrite trust files under the data directory. See docs/security-properties.md for the full posture. - Egress allowlist scope. The skill-set
egress.allowlist applies to the built-inweb_searchandgenerate_imagetools. It does not apply to theexecshell sink (curl/wget run as the host shell), to MCP server children (which open their own outbound connections), or to the LLM HTTP client (which talks to the configured provider directly). Operators wanting hard egress control should run wirken inside a network namespace, restricted-egress container, or with OS-level firewall rules. Seecrates/agent/src/egress.rsfor the documented coverage. - Context management. A per-model context engine trims conversations under token budgets before every LLM call, preferring to drop old tool results before touching user or assistant text. Structured compaction events are written to the session log and projected back into the prompt so the model knows what was trimmed.
- Prompt injection detection. Inbound messages are scanned for role-switching attempts, instruction overrides, base64-encoded commands, tool-call injection, and system prompt extraction. Detected threats are flagged in the session log and forwarded to SIEM; messages are not blocked.
- Confidential inference. Tinfoil and Privatemode providers run LLMs inside hardware enclaves (AMD SEV-SNP, Intel TDX). Prompts are encrypted end-to-end and protected against software attacks on infrastructure. See the Privatemode reference instance for the end-to-end trust model and deployment recipe.
- Encrypted credentials. XChaCha20-Poly1305 vault keyed from the OS keychain. Per-credential expiry and rotation. No plaintext export. MCP credentials are isolated in a separate proxy process.
- Centralized policy.
wirken setup --org https://wirken.corp.example.compulls provider, SIEM, MCP, and permission config from a company endpoint. Developers get grab-and-go setup. IT manages one config. Policy refreshes on everywirken run.
Documentation
- Getting started
- Deploying for teams (shared inference, per-channel adapters, current limitations)
- Permissions and identity (what exists today, what is planned)
- CLI reference
- Configuration reference
- Channel setup (Telegram, Discord, Slack, Teams, Matrix, Signal, Google Chat, iMessage)
- Multi-agent setup
- Skills guide (markdown skills, Wasm skills, registry)
- Lyrik overview (security-assessment skill — what it is and why)
- MCP setup
- Security properties (OWASP and NIST AI RMF mappings)
- Enterprise deployment (org config, SIEM, sandbox)
- Migration from OpenClaw
- Troubleshooting
- Architecture
- Enforcement model (compile-time vs. runtime guarantees)
- Release process (version bump, tag, sign, publish, smoke test)
- Release signing (Ed25519 key, rotation, verification)
Contributing
Wirken is a Rust workspace. All crates compile and test independently:
cargo test # full test suite cargo test -p wirken-vault # test one crate cargo build -p wirken-cli # build the binary
Building from source requires the Cap'n Proto compiler (capnproto package on Ubuntu, capnp via Homebrew on macOS).
The architecture is documented in docs/architecture.md.
Adapter contributions are especially welcome. Each adapter is an independent crate (crates/adapter-<channel>/) that implements the same IPC contract: connect to the gateway UDS, perform Ed25519 handshake, convert platform messages to/from Cap'n Proto frames. See any existing adapter for the pattern (Telegram is the simplest; Teams shows the HTTP webhook variant).
Status
Wirken 0.9.x is the current series. 0.9 gets fixes and features; 0.8 gets security fixes only. 0.7 is unsupported.
- 9 channel adapters under
crates/adapter-*: Telegram, Discord, Slack, Microsoft Teams, Matrix, WhatsApp, Signal, Google Chat, iMessage. - 8 LLM providers in
crates/agent/src/llm.rs: Ollama, Anthropic, OpenAI, Google Gemini, AWS Bedrock, Tinfoil, Privatemode, plus acustomprovider for any OpenAI-compatible endpoint. - 16 bundled skills under
skills/. - 766 tests in the workspace, all green on main (
cargo test --workspace). - Signed releases.
checksums.sha256is signed offline with an Ed25519 SSH key.install.shembeds the public key inline, fetcheschecksums.sha256.sigfrom the release, and fails closed on any verification failure. See docs/release-signing.md and KEYS.
The name
Wirken: German, to work, to weave, to have effect. Named for Gebruder Ottenheimer, a weaving mill in Wurttemberg, 1862-1937.
License
MIT