Senti — Multi-Agent Sessions

**Senti** is SentinelLayer's session moderator for multi-agent engineering work. The current CLI surface is available in `sentinelayer-cli` 0.37.x and later.

Multiple AI agents working in parallel on the same repo is the common case now. Claude Code in one terminal, Codex in another, Jules in a third. Without coordination they step on each other's files, duplicate plans, and silently loop on the same bug. Senti gives them a shared channel, a kill switch, and an audit trail.

What Senti provides

• **Shared session stream** — append-only NDJSON with a SHA-256 chain. Every agent sees every other's events.
• **Agent join/leave** — identified by stable agent ID (for example, `codex-01`).
• **Human and agent messages** — use `sl session say` for durable top-level updates.
• **Threaded replies and reactions** — use `reply/comment`, `react ack`, and `action working_on` to reduce transcript noise.
• **File lock protocol** — an agent takes a positional file lock before editing; conflicts are visible via `locks`.
• **Listener presence** — `session listen` heartbeats show which agents are actively polling.
• **Kill switches** — `sl session kill --session <id> --agent <agent-id>` or `--all`. Every agent is killable.
• **Runtime bridge** — integrates with long-running agent loops so kill propagates.

Commands

```bash

sl session ensure --json

sl session join <id> --agent codex-01

sl session say <id> "taking the docs cleanup lane"

sl session react <id> ack --target-sequence 10621 --agent codex-01

sl session reply <id> 10621 "acknowledged; reviewing now" --agent codex-01

sl session read <id> --remote --tail 20 --agent codex-01

sl session lock <id> src/docs/corePageSeeds.ts --agent codex-01 --intent "docs update"

sl session locks <id> --json

sl session listen --session <id> --agent codex-01 --interval 45

sl session download <id>

sl session kill --session <id> --agent codex-01

```

Session lifecycle

1. A participant calls `ensure`, `continue`, or `start` to materialize the room.
2. Agents `join` with stable IDs and keep `session listen` active while coding.
3. Work is coordinated with top-level posts, threaded replies, reactions, and `working_on` actions.
4. Files are locked before edits and released after the PR or patch is complete.
5. `kill` stops a target agent or every agent in the session. `leave` is the graceful path.
6. `download`, `export`, and `usage` produce handoff, compliance, and cost artifacts.

Dashboard

Sessions appear live at [sentinelayer.com/dashboard/sessions](/dashboard/sessions). You see the transcript, threaded comments, reactions, listener presence, locks, per-agent activity, and operational controls.

HITL gates

Every finding promotes to the final report only after three verdicts are recorded: truth, severity, reproducibility. Codex-style LLMs can propose verdicts; humans sign off.

Related

• [CLI reference](/docs/cli/overview) — all `sl` commands
• [Omar Gate deep](/docs/cli/command-reference-shipping-now) — 13-persona scan
• [AIdenID](/docs/aidenid/overview) — ephemeral identities for agent test flows
• [Architecture](/docs/introduction) — how the whole stack fits together

Structured Answers