A reasoning-heavy model can burn the whole max_tokens budget on
thinking blocks; the parsed turn then has neither text nor tool calls
and the loop treated it as a successful completion, marking a run done
when nothing happened. Such a first turn now fails the run so the
failure is visible and the heartbeat retries; later empty-text turns
after real tool work still complete normally.
The CEO's first run hired five reports into departments "general" and
"engineering"; the hiring-policy lookup was exact-match against the
seeded names (Executive, Engineering, ...), so every hire failed and
the kickoff produced nothing. The tool schema now enumerates the
canonical departments, the policy lookup matches case-insensitively
and stores the canonical name on the agent, and an unknown department
error lists the valid choices so the model can self-correct.
The abort timer was cleared as soon as fetch resolved with headers, so
a server that stalled the response body hung the request forever - and
one hung run froze the autonomous queue behind the worker re-entrancy
guard. JSON requests now carry a header deadline plus a longer body
deadline on one abort controller, and event streams fail when no chunk
arrives within an idle window, cancelling the reader. Timeouts surface
as ProviderHttpError so the run records a clear failure and the
heartbeat retries.
A task_run still marked running at startup has no executor - the
single-instance app that owned it is gone - yet nothing failed it, the
queue worker only claims queued rows, and the heartbeat excludes tasks
with a running run, so one interrupted shutdown wedged the company
forever. Boot now fails orphaned running runs and their in-flight
foreground provider requests (auditably), which hands the task back to
the heartbeat for an automatic retry. Background provider requests are
untouched; the background worker resumes those by external id.
Adversarial review of the kickoff feature surfaced two gaps. First, the
kickoff task was inserted with raw SQL, so a workspace-bound kickoff
carried requires_filesystem without the execution approval every other
task receives - manual runs failed and review-mode orgs had no approval
trail. Kickoffs (in-app and onboarding) now go through createTask and
its policy gates. Second, a goal created before any agent existed
stayed inert forever; creating an agent now operationalizes every
active goal that has no task yet. Goal owners must be active agents:
an archived owner falls back to the top-level agent and the owner
dropdown only offers active agents.
Settings shrinks from 14 tabs to the five that serve the product:
Workspace, Billing, Providers, Skills, Migration. The deleted panels
(Members/IAM, Models, Policies, Operations, SDLC, Compliance, Profile,
Notifications, Security) were read-only governance tables and raw
audit-log dumps with no old-app counterpart. The Automation page goes
too: its rules stored empty conditions and no engine ever executed
them. Sidebar reordered to match the old app hierarchy (Home, Inbox,
Work, Approvals first). Backend services and schema are untouched.
A goal created from the Work page only wrote a goals row; nothing ever
read it, so the company never started working on it. Creating a goal now
produces an "Operationalize goal" kickoff task for the chosen owner (or
the top-level agent when unowned) and re-enables that agent's heartbeat,
so the autonomous queue worker starts the run - the same activation
contract the previous runtime had via goal_activated wakes and CEO
bootstrap. Onboarding shares the same kickoff module instead of its own
copy.
Migration 041 adds parent_task_id to tasks for subtask hierarchy.
TaskDetailDialog now has three tabs: Overview (description, assignee,
project, goal, workspace, runs with error messages), Subtasks (child
tasks with status), and Activity (audit events and approval history).
Inline edit overlay covers title, description, assignee, risk, priority,
and parent task. Task delete is available from the dialog footer.
Every catalog provider can now drive agent execution:
- Azure OpenAI -> Chat Completions at the deployment URL with api-key
- AWS Bedrock -> Converse API signed natively with SigV4 (node:crypto)
- Vertex AI -> Gemini generateContent with a service-account JWT (RS256)
exchanged for a cached access token
The credential resolver maps each cloud profile to the turn's needs; the router
covers all 16 providers. Pure encoders, the SigV4 signer and the GCP auth are
unit-tested; no AWS/GCP SDK dependency.
New 'Custom OpenAI-compatible' and 'Custom Anthropic-compatible' providers let
an agent run against any compatible endpoint by base URL + key: the turn posts
Chat Completions (/chat/completions) or Anthropic Messages (/v1/messages) at the
configured base URL. Backfilled into existing orgs by migration 037.
The loop was gated to API-key providers, so a GPT-account (Codex) CEO failed
with 'requires an API-key provider'. Port Pi's actual request handlers natively
so every agent-capable provider runs:
- chat-completions turn for xAI, Groq, Mistral, Cerebras, OpenRouter
- Codex SSE Responses for GPT account
- Anthropic-OAuth (Claude Code system + beta headers) for Claude Pro/Max
- Copilot proxy (baseUrl from the token) for GitHub Copilot
- Gemini Code Assist for Google Antigravity
A credential resolver refreshes expired OAuth access tokens at request time and
persists the rotation; the runner routes by provider key and passes the resolved
credential. The heartbeat now retries failed tasks (bounded to 3 attempts) so a
run that failed on the old gate recovers automatically. Cloud providers
(Azure/Bedrock/Vertex) still report a clear unsupported message.
The queue worker now also enqueues one run per actionable assigned task of a
heartbeat-enabled agent with an enabled provider and no run yet, so after the
CEO hires and delegates, its reports start their tasks on their own. Each task
is enqueued at most once. The team-building prompt is now scoped to managers
(no manager) with an empty team, so reports execute their task instead.
The new app reached a real LLM but never ran an agent: the kickoff task_run
sat queued forever and runTask was a single text completion with no tools.
This adds the missing runtime:
- provider-tool-turn: OpenAI Responses + Anthropic Messages tool-use turns
- agent-tools + deliverable-writer: hire_agent, create_task,
update_task_status, write_deliverable (workspace-scoped, traversal-safe)
- agent-run-loop: real tool-use loop, capped iterations
- task-queue-worker: claims queued runs in a transaction and executes them
task-execution now drives the loop; a background worker auto-runs the
bootstrap CEO run so it hires its team, delegates tasks, and writes the first
deliverable into the bound workspace. Autonomy is gated by AGENTCOMPANY_NEW_AUTORUN
(tests run with it off to stay hermetic).
Closes the Anthropic OAuth gap with a native Claude Pro/Max device sign-in
(reusing the OAuth engine) and adds three OpenAI-compatible API-key
providers. Model lists are sourced from the provider registry. A backfill
migration seeds the new providers into existing organizations idempotently.
Implement the official OAuth login flows natively (ported from pi-ai's
reference): GitHub Copilot device-code, GPT account (ChatGPT Codex) and
Google Antigravity authorization-code + PKCE via a loopback callback server.
Obtained credentials are stored as JSON through the existing secret vault.
Onboarding and Settings now run the real sign-in instead of pasting an
access token.
ZacharyZhang-NY