Binary reference

ai-gateway

Unified AI gateway for command catalog, policy firewall, and safe execution

Version10.0.3Size10.0MBAuthorWarith Al MaawaliLicenseProprietaryWebsitehttps://www.digi77.com

File Information

Property	Value
Binary Name	ai-gateway
Version	10.0.3
Build Date	Not disclosed
Rust Version	1.82.0
File Size	10.0MB
Author	Warith Al Maawali
License	Proprietary
Category	Kodachi Binary
Description	Unified AI gateway for command catalog, policy firewall, and safe execution
Git Commit	unknown
Metadata Generated	2026-11-13T16:19:42Z
Binary Timestamp	Unknown
JSON Data	View Raw JSON

SHA256 Checksum

aa8f3124041312715c94085096eabdee62dfab06b1c5a1b2c41580605fe5947e

Features

#	Feature
1	Embedded command catalog with 606 commands from 26 services
2	TF-IDF + substring hybrid search across all services
3	Three-tier risk classification (Passive/Active/Dangerous)
4	Policy firewall with allowlist enforcement
5	Workspace confinement and path validation
6	Failure cooldown to prevent cascading degradation
7	Per-agent identity, rate limiting, and audit trail for executions
8	Approval ticket workflow for agent-to-human authorization
9	Configurable per-operation timeouts with env var override
10	Full JSON/JSON-pretty/JSON-human output support

Security Features

Feature	Description
Input Validation	Argument parsing via clap; per-command validation is the consumer's responsibility
Rate Limiting	Not provided by cli-core
Authentication	Not provided by cli-core (see online-auth)
Encryption	Not provided by cli-core

System Requirements

Requirement	Value
OS	Linux (Debian-based)
Privileges	root/sudo for system operations
Dependencies	OpenSSL, libcurl

Global Options

Flag	Description
`-h, --help`	Print help information
`-v, --version`	Print version information
`-n, --info`	Display detailed information
`-e, --examples`	Show usage examples
`--json`	Output in JSON format
`-o, --output-format <FORMAT>`	Force output format (text\|json)
`--json-pretty`	Pretty-print JSON output with indentation
`--json-human`	Enhanced JSON output with improved formatting (like jq)
`--json-filter <FIELD1,FIELD2>`	Return only specified JSON fields (comma-separated)
`--fields <FIELD_LIST>`	Select specific fields to include in output (comma-separated)
`--offset <NUMBER>`	Skip first N results (for pagination)
`--port <PORT>`	Set custom port number (1024-65535)
`--log-level <LEVEL>`	Set log level (error\|warn\|info\|debug)
`--verbose`	Enable verbose output
`--quiet`	Suppress non-essential output
`--no-color`	Disable colored output
`--config <FILE>`	Use custom configuration file

Commands

Administration

`index`

Rebuild the search index from the embedded registry

Usage:

ai-gateway index rebuild

Options:

rebuild: Rebuild the TF-IDF search index over the command catalog and print the service and command counts. Run this after upgrading the hook binaries.

Examples:

ai-gateway index rebuild

ai-gateway index rebuild --json

ai-gateway index rebuild --json-pretty

`policy`

Show the current safety policy configuration

Usage:

ai-gateway policy show

Options:

show: Display the policy firewall: risk-tier rules, the service allowlist, and whether KODACHI_PENTEST_MODE is currently enabled. Read-only.

Examples:

ai-gateway policy show

ai-gateway policy show --json

ai-gateway policy show --json-pretty

`doctor`

Health check that verifies all registered binaries exist and respond

Usage:

ai-gateway doctor

Options:

(no command-specific options): doctor probes every registered binary and reports whether it exists, responds, and at which version. Use the global --json, --json-pretty or --json-human to pick the output shape.

Examples:

ai-gateway doctor

ai-gateway doctor --json

ai-gateway doctor --json-pretty

`approve`

Issue or verify human approval tickets for agent execution

Usage:

ai-gateway approve issue <service> <command> [--agent-id <id>] [--ttl <secs>]
ai-gateway approve verify --ticket <ticket> <service> <command> [--agent-id <id>]

Options:

issue <SERVICE> <COMMAND>: Issue a signed, time-limited approval ticket for one service and command (requires KODACHI_GATEWAY_APPROVAL_SECRET)
verify --ticket <TICKET> <SERVICE> <COMMAND>: Verify a ticket against the service and command it was issued for
--agent-id <ID>: Agent identity the ticket is bound to
--ttl <SECS>: Ticket lifetime in seconds, 1 to 3600 (issue only)

Examples:

ai-gateway approve issue health-control block-internet --agent-id nullclaw --ttl 600 --json

ai-gateway approve verify --ticket "$TICKET" health-control block-internet --agent-id nullclaw --json

Discovery

`list`

List all services and their commands from the embedded registry

Usage:

ai-gateway list [--service <name>] [--category <name>]

Options:

--service: Filter by service name
--category: Filter by category name

Examples:

ai-gateway list

ai-gateway list --json

ai-gateway list --service tor-switch

ai-gateway list --service health-control --json

ai-gateway list --category passive

`search`

Search across all services and commands using TF-IDF + substring matching

Usage:

ai-gateway search <query> [--limit <N>]

Options:

--limit: Maximum number of results to return

Examples:

ai-gateway search "tor status"

ai-gateway search "dns leak" --json

ai-gateway search "network check" --limit 5

ai-gateway search "panic" --json-pretty

`help`

Show detailed help for a service or specific command

Usage:

ai-gateway help <service> [command]

Options:

<SERVICE>: Service name from the embedded catalog, for example tor-switch. An --agent-id value such as nullclaw is not a service and returns "Service not found".
[COMMAND]: Optional command name. Omit it for the whole service spec.

Examples:

ai-gateway help tor-switch

ai-gateway help tor-switch tor-status --json

ai-gateway help health-control --json-pretty

ai-gateway help dns-leak test

ai-gateway help ip-fetch fetch --json

`capabilities`

Show effective capabilities for an AI agent identity

Usage:

ai-gateway capabilities [--agent-id <id>] [--agent-token <token>]

Options:

--agent-id: Agent profile (nullclaw/agentzero/openclaw/picoclaw/nanoclaw/claude-code/gpt/gemini/open-interpreter; aliases: nanclaw, "claude code", "open interpreter")
--agent-token: Optional token for authenticated agent profiles

Examples:

ai-gateway capabilities --agent-id nullclaw --agent-token $KODACHI_AGENT_TOKEN_NULLCLAW --json

ai-gateway capabilities --agent-id nullclaw --agent-token $KODACHI_AGENT_TOKEN_NULLCLAW

ai-gateway capabilities --agent-id openclaw --agent-token $KODACHI_AGENT_TOKEN_OPENCLAW --json

ai-gateway capabilities --agent-id picoclaw --agent-token $KODACHI_AGENT_TOKEN_PICOCLAW --json

ai-gateway capabilities --agent-id nanoclaw --agent-token $KODACHI_AGENT_TOKEN_NANOCLAW --json

ai-gateway capabilities --agent-id nanclaw --agent-token $KODACHI_AGENT_TOKEN_NANOCLAW --json

ai-gateway capabilities --agent-id claude-code --agent-token $KODACHI_AGENT_TOKEN_CLAUDE_CODE --json

ai-gateway capabilities --agent-id "claude code" --agent-token $KODACHI_AGENT_TOKEN_CLAUDE_CODE --json

ai-gateway capabilities --agent-id gpt --agent-token $KODACHI_AGENT_TOKEN_GPT --json

ai-gateway capabilities --agent-id gemini --agent-token $KODACHI_AGENT_TOKEN_GEMINI --json

ai-gateway capabilities --agent-id open-interpreter --agent-token $KODACHI_AGENT_TOKEN_OPEN_INTERPRETER --json

ai-gateway capabilities --agent-id "open interpreter" --agent-token $KODACHI_AGENT_TOKEN_OPEN_INTERPRETER --json

ai-gateway capabilities --agent-id anonymous --json

Execution

`run`

Validate and execute a binary command through the safety layer

Usage:

ai-gateway run <service> --command <cmd> [--args <args>...] [--args-json <json>] [--dry-run] [--timeout <secs>] [--agent-id <id>] [--agent-token <token>] [--approval-ticket <ticket>] OR ai-gateway run --batch-json <json> --agent-id <id> --agent-token <token>

Options:

<SERVICE>: Service to invoke (required unless --batch-json is used)
--command <CMD>: Command to execute (required unless --batch-json is used)
--args: Command arguments (multiple)
--args-json: Command arguments as JSON object/array
--batch-json: Batch execution payload for trusted agents (JSON array, sequential mode only)
--dry-run: Validate without executing
--timeout: Execution timeout in seconds
--confirm: Approval token for commands requiring confirmation
--approval-ticket: Human-issued approval ticket for Active/Dangerous commands
--agent-id: Agent profile (trusted profiles require token auth)
--agent-token: Agent auth token for token-required profiles

Examples:

ai-gateway run health-control --command net-check --dry-run

ai-gateway run ip-fetch --command fetch --dry-run --json

ai-gateway run tor-switch --command tor-status --timeout 60 --json

ai-gateway run tor-switch --command set-exit-node --args-json '["de"]' --dry-run --json

ai-gateway run dns-leak --command test --dry-run --json-pretty

ai-gateway run tor-switch --command tor-status --agent-id claude-code --agent-token "$KODACHI_AGENT_TOKEN_CLAUDE_CODE" --dry-run --json

KODACHI_TRUSTED_BATCH_MODE=true ai-gateway run --agent-id nullclaw --agent-token "$KODACHI_AGENT_TOKEN_NULLCLAW" --batch-json '[{"service":"ip-fetch","command":"fetch","dry_run":true},{"service":"tor-switch","command":"tor-status","dry_run":true}]' --json

KODACHI_PENTEST_MODE=true ai-gateway run health-control --command wipe-file --args-json '{"file":"results/example-to-wipe.txt"}' --confirm "I understand" --dry-run --json

Operational Scenarios

Scenario-oriented workflows generated from the binary's built-in -e --json examples.

Scenario 1: Quick Start

The gateway is the one safe door into every other Kodachi binary: find a command, read its spec, preview it, then run it. ai-gateway itself never needs root, but the service it launches may.

Step 1: Find the command you want without knowing which service owns it

ai-gateway search "tor exit node" --limit 5 --json

Expected Output: Top 5 ranked matches, each with service, command, description and score

Step 2: Read the full spec of a command before you invoke it

ai-gateway help tor-switch tor-status --json

Expected Output: JSON command spec with options, examples and risk tier

Step 3: Preview the invocation: policy check, no execution

ai-gateway run tor-switch --command tor-status --dry-run --json

Expected Output: Validation result showing the resolved binary and argv that would run

Step 4: Run it for real once the dry-run looks right

ai-gateway run tor-switch --command tor-status --json

Expected Output: The live tor-status output, wrapped in the gateway envelope

Note

Passive commands need no confirmation and no env var. Active commands need --confirm, and Dangerous commands additionally need KODACHI_PENTEST_MODE=true.

Scenario 2: Service Discovery

Browse the embedded catalog of every Kodachi service and command. The catalog is compiled into the binary, so list works offline and without authentication.

Step 1: List every service in the catalog with its commands

ai-gateway list --json

Expected Output: JSON array of the registered services, each with its command list and metadata

Step 2: Narrow the catalog to one service

ai-gateway list --service tor-switch --json

Expected Output: The tor-switch commands with descriptions and risk tiers

Step 3: Filter by command category instead of by service

ai-gateway list --category Discovery

Expected Output: The commands whose category matches, across all services

Note

--category (-c) matches the category name a service assigns to its own commands, so the useful values differ per service. Run 'ai-gateway list --service <name>' first to see which categories that service uses.

Step 4: Same listing in readable text instead of JSON

ai-gateway list --service health-control

Expected Output: A formatted list of the health-control commands

Scenario 3: Command Search

Search across all services using TF-IDF + substring hybrid search engine

Step 1: Search for Tor exit node commands

ai-gateway search "tor exit node" --json

Expected Output: Ranked results with relevance scores

Note

Uses TF-IDF cosine similarity + substring matching

Step 2: Search with limited results

ai-gateway search "dns leak" --limit 5 --json

Expected Output: Top 5 matching commands for DNS leak operations

Step 3: Search for network check commands in text format

ai-gateway search "network check"

Expected Output: Numbered list of matching commands with scores

Step 4: Find all panic/emergency commands

ai-gateway search "panic" --json

Expected Output: Emergency commands with Dangerous risk tier

Note

Dangerous commands are blocked unless pentest mode

Scenario 4: Service Help

Get detailed help for any service or specific command

Step 1: Full specification for tor-switch service

ai-gateway help tor-switch --json

Expected Output: JSON with all commands, options, examples, features, requirements

Step 2: Help for a specific command

ai-gateway help health-control security-status --json

Expected Output: Detailed command spec with parameters and examples

Step 3: Text help for dns-switch service

ai-gateway help dns-switch

Expected Output: Formatted help output for dns-switch

Step 4: Pretty-printed JSON help for ip-fetch

ai-gateway help ip-fetch --json-pretty

Expected Output: Indented JSON with full service specification

Scenario 5: Safe Command Execution

Validate first, then execute when appropriate. No env var is needed for passive/active commands.

Step 1: Step 1 (validate): preview a passive command

ai-gateway run tor-switch --command tor-status --dry-run --json

Expected Output: Validation result without actual execution

Note

Always use --dry-run first for safety

Step 2: Step 2 (apply): execute passive command (no env needed)

ai-gateway run tor-switch --command tor-status --json

Expected Output: Live command execution result

Note

Passive commands execute without KODACHI_PENTEST_MODE

Step 3: Active command flow (no env needed): validate with explicit confirmation

ai-gateway run tor-switch --command set-exit-node --args-json '["de"]' --confirm --dry-run --json

Expected Output: Validated active operation with parsed args-json

Note

For active commands, remove --dry-run to apply changes

Step 4: Dry-run DNS leak test with custom timeout

ai-gateway run dns-leak --command test --dry-run --timeout 60 --json

Expected Output: Validation result for dns-leak test

Note

Timeout can be overridden with KODACHI_TOOL_TIMEOUT_MS env var

Step 5: Dry-run ip-fetch fetch command

ai-gateway run ip-fetch --command fetch --dry-run --json

Expected Output: Validation result for ip-fetch fetch

Note

Dry-run keeps examples deterministic in offline/auth-restricted environments

Step 6: Pass structured JSON arguments for agent-safe invocation

ai-gateway run tor-switch --command set-exit-node --args-json '["de"]' --dry-run --json

Expected Output: Dry-run validation with args expanded from JSON

Note

--args-json avoids brittle shell quoting in agent pipelines

Step 7: Pass plain positional arguments straight through to the service

ai-gateway run tor-switch --command set-exit-node --args de --dry-run --json

Expected Output: Dry-run validation showing the argv the service would receive, including the passed-through arguments

Note

--args (-a) takes any number of values and forwards them verbatim, so everything after it is consumed as an argument until the next flag. Use --args-json instead when the values could be mistaken for gateway flags.

Step 8: Run under an agent identity so the call is rate-limited and audited

ai-gateway run health-control --command net-check --agent-id claude-code --agent-token "$KODACHI_AGENT_TOKEN_CLAUDE_CODE" --dry-run --json

Expected Output: Dry-run validation plus the identity that was authenticated and the risk tier it is allowed to reach

Note

The token is read server-side from the env var KODACHI_AGENT_TOKEN_<AGENT_ID>, uppercased with '-' replaced by '_'. It must be EXPORTED: if it is unset, the shell expands it to nothing and clap fails with "a value is required".

Scenario 6: Safety Policy

Read the policy firewall: which services are allowlisted, how commands are classified into the Passive, Active and Dangerous risk tiers, and what each tier demands before it will run.

Step 1: Show the active safety policy as JSON

ai-gateway policy show --json

Expected Output: The risk-tier rules, the service allowlist, and whether pentest mode is currently on

Step 2: Show the same policy as readable text

ai-gateway policy show

Expected Output: A formatted policy summary

Step 3: Indented policy JSON, easier to read while debugging a block

ai-gateway policy show --json-pretty

Expected Output: Indented JSON with the complete policy details

Note

If a run was refused, this is the command that tells you why: check the command's tier and whether its service is on the allowlist.

Step 4: Preview a Dangerous command: it needs the env var AND an explicit confirmation token

KODACHI_PENTEST_MODE=true ai-gateway run health-control --command wipe-file --args-json '{"file":"results/example-to-wipe.txt"}' --confirm "I understand" --dry-run --json

Expected Output: Validation of the dangerous invocation, with nothing executed because of --dry-run

Note

KEEP --dry-run. The example target is only illustrative. Without dry-run this really does wipe the target file, and wipe-file is irreversible. KODACHI_PENTEST_MODE=true is required only for the Dangerous tier, and --confirm must be exactly "I understand" or dangerous:<service>:<command>.

Scenario 7: Administration and Troubleshooting

Keep the search index fresh and find out why a service will not run. These are the commands to reach for when a run fails and you do not know whether the fault is policy, the index, or a missing binary.

Step 1: Rebuild the TF-IDF search index over the command catalog

ai-gateway index rebuild --json

Expected Output: Index statistics: how many services and commands were indexed

Note

Do this after upgrading the hook binaries, otherwise search still ranks against the previous catalog.

Step 2: Probe every registered binary and report whether it is usable

ai-gateway doctor --json

Expected Output: Per-binary health: whether the file exists, whether it responds, its version and its size

Note

This is what to run when a run failed with "Binary not found": doctor shows exactly which entries have no binary behind them.

Step 3: Same health check as a readable table

ai-gateway doctor

Expected Output: A table of binary names against their health status

Note

doctor executes each binary with a probe flag, so it is slower than the other commands here.

Scenario 8: Output Formats and Metadata

Every command accepts the same three JSON shapes, plus the binary's own metadata flags. Pick compact JSON for scripts, pretty for reading, and human when you want it formatted like jq.

Step 1: Compact JSON: one line, best for piping into a script

ai-gateway doctor --json

Expected Output: A single-line JSON envelope

Step 2: Indented JSON for reading a result by eye

ai-gateway search "dns leak" --json-pretty

Expected Output: The same envelope, indented across multiple lines

Step 3: Formatted, jq-style JSON

ai-gateway policy show --json-human

Expected Output: The policy envelope, highlighted and formatted for a human

Step 4: Report what this build of the gateway can do

ai-gateway --info --json

Expected Output: JSON with the feature list, security properties and requirements

Step 5: Print the gateway version

ai-gateway --version

Expected Output: The ai-gateway version string

Scenario 9: AI Agent Integration

The loop an AI agent runs: search for the command, read its spec, dry-run it, then execute under an agent identity. IMPORTANT: an agent name (nullclaw, agentzero, openclaw, picoclaw, nanoclaw, claude-code, gpt, gemini, open-interpreter) is a value for --agent-id, NOT a service you can 'run'. The service argument of 'run' must be a Kodachi binary from the catalog, such as tor-switch or ip-fetch. Use 'ai-gateway list' to see what is runnable.

Step 1: Step 1: the agent finds the best matching command

ai-gateway search "check tor status" --limit 1 --json

Expected Output: The single top match, with its service, command and score

Note

ai-cmd delegates its execution to ai-gateway, so an agent that goes through the gateway gets the same safety checks either way.

Step 2: Step 2: the agent reads the machine-readable command spec

ai-gateway help tor-switch tor-status --json

Expected Output: The command spec: parameters, defaults, examples and risk tier

Note

'help' resolves the name against the embedded service catalog. An --agent-id value such as nullclaw is not a service, so 'help nullclaw' returns "Service not found".

Step 3: Step 3: the agent validates the invocation before committing to it

ai-gateway run tor-switch --command tor-status --dry-run --json

Expected Output: Validation passes and shows exactly what would have run

Step 4: Step 4: the agent executes under its own identity, so the call is rate-limited and audited

ai-gateway run tor-switch --command tor-status --agent-id claude-code --agent-token "$KODACHI_AGENT_TOKEN_CLAUDE_CODE" --json

Expected Output: The live tor-status result, plus the identity and rate limit that were applied

Note

Every named profile except 'anonymous' requires a token. Export KODACHI_AGENT_TOKEN_CLAUDE_CODE first: if it is unset the shell drops the word and clap fails with "a value is required".

Step 5: Ask the gateway what a given agent identity is actually allowed to do

ai-gateway capabilities --agent-id nullclaw --agent-token "$KODACHI_AGENT_TOKEN_NULLCLAW" --json

Expected Output: The effective services, the highest risk tier allowed, and the per-minute rate limit

Note

This is the correct way to work with an agent name. 'run nullclaw' is not: nullclaw is an identity, and it is only a runnable service if you have separately installed a third-party binary of that name (see the External agent runtimes category).

Scenario 10: External Agent Runtimes

Kodachi does not ship any agent runtime binary. The policy allowlist reserves two optional third-party names (nullclaw, aeroclaw) and the registry discovers them at runtime, but only if you have installed a binary of that name into the hooks directory yourself. On a stock install none of them exists, so running one fails with "Binary not found". Everything below assumes you installed it.

Step 1: Check whether an external agent runtime is actually installed

ai-gateway doctor --json

Expected Output: The per-binary health report. An external runtime you have not installed simply does not appear, which is the normal state.

Note

Do this before you write a script against nullclaw or aeroclaw. They are optional and are not part of a stock Kodachi install.

Step 2: Confirm which service names the policy firewall will even accept

ai-gateway policy show --json

Expected Output: The allowlist, which includes the Kodachi services plus the reserved external names nullclaw and aeroclaw

Note

Being on the allowlist only means the policy will not refuse the name. The binary still has to exist on disk. Names such as agentzero, openclaw, picoclaw and nanoclaw are NOT on the allowlist: they exist only as --agent-id identities.

Step 3: List what an installed external runtime exposes

ai-gateway list --service nullclaw --json

Expected Output: The commands the runtime reported when the registry probed it, or a not-found error if it is not installed

Note

The registry discovers an external runtime by executing it with a probe flag under a 5 second timeout, so its command list comes from the binary itself, not from the compiled catalog.

Scenario 11: Agent Identity, Tokens and Approval Tickets

Every execution can carry an agent identity, which decides the highest risk tier it may reach, its rate limit, and what lands in the audit trail. Valid --agent-id values: anonymous, nullclaw, agentzero, openclaw, picoclaw, nanoclaw, claude-code, gpt, gemini, open-interpreter. TOKENS: every profile except anonymous requires one. The gateway reads it server-side from the env var KODACHI_AGENT_TOKEN_<AGENT_ID>, uppercased with '-' turned into '_', and that variable must be EXPORTED before you run these commands. If it is unset, the shell expands $KODACHI_AGENT_TOKEN_... to nothing, the word disappears from the command line, and clap fails with "a value is required".

Step 1: Set the token for an agent identity, once per shell

export KODACHI_AGENT_TOKEN_CLAUDE_CODE="<your-token>"

Expected Output: No output. Every later --agent-id claude-code call can now authenticate.

Note

The env var name is derived from the agent id: nullclaw becomes KODACHI_AGENT_TOKEN_NULLCLAW, open-interpreter becomes KODACHI_AGENT_TOKEN_OPEN_INTERPRETER.

Step 2: Ask what a trusted agent identity is allowed to do

ai-gateway capabilities --agent-id nullclaw --agent-token "$KODACHI_AGENT_TOKEN_NULLCLAW" --json

Expected Output: Its effective services, its maximum risk tier, and its per-minute rate limit

Step 3: Show the default capabilities of an unidentified caller

ai-gateway capabilities --agent-id anonymous --json

Expected Output: The anonymous profile: no token needed, and a lower rate limit than the named profiles

Note

anonymous is the only profile that needs no token, and it is the default when you omit --agent-id.

Step 4: A human issues a time-limited ticket authorizing one agent to run one command

ai-gateway approve issue health-control block-internet --agent-id nullclaw --ttl 600 --json

Expected Output: A signed approval ticket string, valid for the TTL you gave

Note

Requires KODACHI_GATEWAY_APPROVAL_SECRET to be exported. --ttl is in seconds and must be 1 to 3600 (default 300). The ticket is bound to that exact service, command and agent id.

Step 5: Check a ticket is genuine and still valid before relying on it

ai-gateway approve verify --ticket "$TICKET" health-control block-internet --agent-id nullclaw --json

Expected Output: Whether the ticket verifies against this service, command and agent, and when it expires

Note

The service and command are positional and must match the ones the ticket was issued for, or verification fails.

Step 6: The agent presents its identity and its ticket to run a gated command

ai-gateway run health-control --command block-internet --agent-id nullclaw --agent-token "$KODACHI_AGENT_TOKEN_NULLCLAW" --approval-ticket "$TICKET" --dry-run --json

Expected Output: Execution gated by the identity policy, the rate limit and ticket verification. With --dry-run, nothing runs.

Note

block-internet cuts all network access on this machine. Drop --dry-run only when you mean it. Without a valid ticket the gateway refuses and returns requires_approval.

Step 7: Run several invocations in one call, sequentially, as a trusted agent

KODACHI_TRUSTED_BATCH_MODE=true ai-gateway run --agent-id nullclaw --agent-token "$KODACHI_AGENT_TOKEN_NULLCLAW" --batch-json '[{"service":"ip-fetch","command":"fetch","dry_run":true},{"service":"tor-switch","command":"tor-status","dry_run":true}]' --json

Expected Output: One result per entry in the array, in order, each with its own policy verdict

Note

Batch mode needs three things at once: KODACHI_TRUSTED_BATCH_MODE=true, an authenticated agent, and a profile that requires a token. anonymous can never batch. When --batch-json is used, the service argument and --command are supplied inside the JSON instead of on the command line. Each entry may set its own dry_run.

Environment Variables

Variable	Description	Default	Values
`NO_COLOR`	Disable all colored output when set	unset	1\|true\|yes (any value disables color)

Exit Codes

Code	Description
0	Success
1	General error (with --json, the error's code field names its kind)
2	Invalid arguments
78	This CPU has no AVX2 support, which the Kodachi AI tools need