— What FerrFleet does.
We don't reimplement the model, tool use, or context management — Claude Code does that. FerrFleet catalogs your agents, runs them, and shows you what happened.
Each agent is a versioned Claude Code definition: a prompt, a working directory, the sub-agents it can spawn, and the connectors it can reach. Create, edit, and fork agents from the dashboard; every run is launched from a catalog entry.
name: flow-release-bot working_dir: /repo prompt: | Draft release notes from merged PRs since the last tag and open a PR.
Run an agent on demand from the dashboard, on a cron schedule, or from a signed webhook. Schedules are org-scoped and each fire launches a fresh run. The webhook secret is the only credential and is compared in constant time.
# schedule
cron: "0 9 * * 1" # every Monday 09:00
# or a signed webhook
POST /api/v1/schedules/{id}/trigger
x-ferrfleet-webhook-secret: <secret>Every run records tokens in, tokens out, and the compute cost we derive from them, attributed to the agent that ran it. The Spend page rolls that up by day and by agent so you can see where the tokens go.
GET /api/v1/metrics/spend?days=14 → flow-release-bot 31 runs $2.54 → track-triager 412 runs $0.82 → vault-rotator 96 runs $1.73
Tail a run as it happens over Server-Sent Events: prompts, tool calls, and assistant output stream straight into the dashboard. Re-open any past run to read the full stored transcript and the session id Claude Code returned.
GET /api/v1/runs/{id}/stream (text/event-stream)
[assistant] Reading merged PRs since v3.1.0…
[tool_use] Bash: git log --oneline
[assistant] Drafting CHANGELOG entries…Give an agent typed access to MCP servers and the rest of the FerrLabs suite. Connector secrets are resolved per run from Vault and injected into the runner — they are never stored in the agent definition or the transcript.
connector: vault-prod kind: mcp # secrets resolved per run, scoped to the # Kubernetes Job that runs the agent
A catalogue of agents that drive the FerrLabs suite — vault-rotator, track-triager, flow-release-bot, growth-seo-auditor, dependency-pr-bot. When your org has an active subscription on the target product, running them is free and does not count against your run quota.
vault-rotator → FerrVault track-triager → FerrTrack flow-release-bot → FerrFlow growth-seo-auditor → FerrGrowth