Configuration¶

Push reads TOML from ~/.push/config.toml by default. Pass --config <path> to use a different file for a gateway, doctor, init, or job command.

push doctor
push

Paths beginning with ~ are expanded. Invalid values, unknown fields inside provider sections, unsafe path overlap, and removed gateway permission settings fail configuration load with an actionable error.

Create the one assistant repository and persist its root before editing the rest of the config:

push init ~/Code/assistant

For a new file, init writes a private, owner-only Telegram and Codex starting point with empty telegram.bot_token and telegram.allow_user_ids values. Fill both in before running Push. Push derives SOUL.md, context/, evals/, and jobs/ from assistant_root. At run time it appends their resolved absolute locations to the user-owned SOUL.md instructions in memory. It does not write machine paths into the repository.

Root configuration, route, and primary-delivery tables do not yet reject every unknown key. Use the documented names, then run push doctor; do not assume a silent key changed runtime behavior.

Minimal configuration¶

channel = "telegram"
agent = "codex"
assistant_root = "~/Code/assistant"

[telegram]
bot_token = "token-from-BotFather"
allow_user_ids = [123456789]

channel is the easiest single-provider setup. agent is claude, codex, or pi. Push preserves backend permission settings for chats. Codex and Claude jobs bypass interactive permissions so unattended runs can complete.

Pi setup¶

Install Pi from pi.dev and configure a model provider or complete its authentication as the same user that runs Push. Confirm pi --version works in the service environment, then select it:

agent = "pi"

Push finds pi through PATH, runs pi --print --mode json, and stores the session ID from Pi's JSON event stream. It resumes the session with --session. Clearing a conversation discards that mapping, so the next turn creates a fresh Pi session. Push appends SOUL.md as system instructions, separate from the user message. Pi is not required unless the default backend, an enabled route, or jobs_agent selects it.

Channels¶

iMessage¶

[imessage]
db_path = "~/Library/Messages/chat.db"
self_handles = ["you@icloud.com"]
allow_from = ["+15551234567"]

At least one self_handles or allow_from value is required when iMessage is enabled. See the iMessage guide.

Telegram¶

[telegram]
bot_token = "token-from-BotFather"
allow_user_ids = [123456789]
allow_chat_ids = []

At least one stable numeric user or chat ID is required. Keep the config file private. push init creates new config files with mode 0600 on Unix. Set TELEGRAM_BOT_TOKEN when an environment variable is a better fit. See the Telegram guide.

Slack¶

[slack]
allow_user_ids = ["U012ABCDEF"]

Slack requires both SLACK_APP_TOKEN and SLACK_BOT_TOKEN, or the matching slack.app_token and slack.bot_token values in the private config. At least one exact Slack member ID is required. See the Slack guide.

Telegram voice notes are optional. Configure the shared voice provider with:

[voice]
openai_api_key = "your-api-key"
name = "cedar"

OPENAI_API_KEY remains available as a higher-priority override for CI and service secret injection. voice.name is optional and defaults to cedar. Supported names are alloy, ash, ballad, coral, echo, fable, nova, onyx, sage, shimmer, verse, marin, and cedar. Without either API key value, text remains fully available and voice notes get a helpful fallback. See Voice Messages.

Run both providers¶

Use channels instead of channel:

channels = ["imessage", "telegram", "slack"]
agent = "codex"

[imessage]
self_handles = ["you@icloud.com"]

[telegram]
bot_token = "token-from-BotFather"
allow_user_ids = [123456789]

[primary_delivery]
channel = "telegram"
target = "123456789"

Each provider polls independently, keeps its own cursor, and replies through the channel and exact conversation that originated the message. Failure in one provider does not stop the other.

primary_delivery is the destination for scheduled job results. The channel must be enabled and the target must appear in that channel's allowlist. Telegram topic targets use "<chat-id>:<topic-id>". Slack primary targets use an allowlisted member ID such as U012ABCDEF.

Routing¶

Routes can override the backend for a channel or exact thread:

[[routes]]
channel = "telegram"
agent = "codex"

[[routes]]
thread = "telegram:dm:123456789"
agent = "claude"

Precedence is:

exact thread or topic route
parent Telegram private-chat route for a topic
channel route
root agent

Thread keys are:

imessage:self:<handle>
imessage:dm:<handle>
telegram:dm:<chat-id>
telegram:dm:<chat-id>:topic:<topic-id>
slack:dm:<workspace-id>:<dm-channel-id>

Agent permissions¶

For chats, Push invokes Claude Code, Codex, and Pi without overriding their sandbox, approval mode, or tool lists. Codex and Claude jobs bypass interactive permissions because scheduled work has no operator available to approve requests. Review permissions and security before enabling jobs.

Settings reference¶

Core¶

Setting	Default	Purpose
`channel`	`"imessage"`	Single enabled provider when `channels` is empty
`channels`	`[]`	Concurrent enabled providers
`agent`	`"claude"`	Default backend
`poll_interval`	`"3s"`	Delay between channel polls
`run_timeout`	`"10m"`	Maximum chat backend run time

iMessage¶

Setting	Default	Purpose
`imessage.db_path`	`~/Library/Messages/chat.db`	Messages database read by the macOS channel
`imessage.self_handles`	`[]`	Own handles accepted in one-to-one self chats
`imessage.allow_from`	`[]`	Other trusted one-to-one sender handles

Telegram¶

Setting	Default	Purpose
`telegram.bot_token`	`TELEGRAM_BOT_TOKEN` fallback	Private Bot API token; the environment value is used when this is omitted
`telegram.allow_user_ids`	`[]`	Trusted numeric sender IDs
`telegram.allow_chat_ids`	`[]`	Trusted numeric private-chat IDs

Slack¶

Setting	Default	Purpose
`slack.app_token`	`SLACK_APP_TOKEN` fallback	App-level Socket Mode token with `connections:write`
`slack.bot_token`	`SLACK_BOT_TOKEN` fallback	Bot token used for `auth.test`, replies, and progress
`slack.allow_user_ids`	`[]`	Trusted stable Slack member IDs

Voice¶

Setting	Default	Purpose
`voice.openai_api_key`	`OPENAI_API_KEY` fallback	Optional key for Telegram transcription and spoken replies; the environment value has priority
`voice.name`	`"cedar"`	OpenAI voice used for spoken replies

Delivery and routes¶

Setting	Default	Purpose
`primary_delivery.channel`	none	Enabled channel used for scheduled job results
`primary_delivery.target`	none	Allowlisted destination on the primary channel
`routes[].thread`	none	Exact channel-qualified thread key to match
`routes[].channel`	none	Enabled provider to match when no exact thread route wins
`routes[].agent`	required per route	Backend selected by the matching route

Local state¶

Setting	Default	Purpose
`assistant_root`	required for new setups	Canonical root of the one assistant repository; `SOUL.md`, `context/`, `evals/`, and `jobs/` are derived
`state_path`	`~/.push/state.json`	Channel cursors and backend session IDs
`database_path`	`~/.push/push.db`	Canonical conversation, approval, and job history
`audit_log_path`	`~/.push/audit.jsonl`	Structured local audit log
`audit_log_content`	`false`	Include message and reply content in audit events

Jobs¶

Setting	Default	Purpose
`jobs_agent`	root `agent`	Default jobs backend
`jobs_max_timeout`	`"30m"`	Maximum accepted job timeout
`jobs_run_dir`	`~/.push/run`	Local advisory locks
`jobs_max_workers`	`2`	Concurrent scheduled job workers

Push validates that runtime state, locks, external config files, and job work directories do not overlap in unsafe ways. Runtime state and secrets must stay outside the Git-versioned assistant repository.

Complete example¶

channels = ["imessage", "telegram"]
agent = "codex"
assistant_root = "~/Code/assistant"
poll_interval = "3s"
run_timeout = "10m"

[imessage]
self_handles = ["you@icloud.com"]
allow_from = []

[telegram]
bot_token = "token-from-BotFather"
allow_user_ids = [123456789]

[primary_delivery]
channel = "telegram"
target = "123456789"

[[routes]]
thread = "telegram:dm:123456789"
agent = "claude"

Legacy flat channel fields remain accepted for migration, but new configurations should use [imessage] and [telegram]. JSON configuration and gateway permission fields are no longer supported. Configure permissions in the selected agent instead.

Push resolves claude, codex, and pi through the service PATH. Configure the backend's model in that backend rather than in Push. Telegram environment tokens always use TELEGRAM_BOT_TOKEN; the iMessage reply marker is internal.

Legacy assistant_dir and jobs_dir settings remain compatible only when the jobs path is exactly <assistant_dir>/jobs. For separate legacy paths, move SOUL.md, context, and jobs under one directory and replace both settings with assistant_root.