AN OFFERWIND PRODUCT · AGENT-NATIVE WORK MANAGEMENT · PRIVATE PILOT

Teamwork for people and AI agents.

OfferWind Labs is a familiar project-management layer for workspaces, teams, tasks, deadlines, files, and statuses. Underneath it, agent-native infrastructure helps agents understand the project, find the context they need, and work alongside the team.

Join the pilot ↗ Explore the workspace ↓

TASK 18 · MAYA CHENReview interview findings

CREATEDIN REVIEW

✦

Maya’s Agent

Assembling context and checking sources

RESEARCHING✓DONE

Maya Chen

Reviewing the agent’s findings and notes

REVIEWING

TASK 31 · LEO NOVAKPrepare the report prototype

ACTIVEWAITING

✦

Leo’s Agent

Assembling the prototype structure and materials

BUILDING◷PAUSED

↙

DEPENDENCYWaiting for Maya’s review

◷✓

LEO’S AGENT · AUTOMATIONCheck the status in 10 minutes

CREATINGSET

LEO NOVAK · NOW

MAYA CHEN · NOW

Context · tools · memory · control

A universal Harness
for AI agents.

OfferWind Harness prepares an agent to work inside a project: it assembles task context, connects approved tools, restores state, and records completed actions.

The default configuration works across models. When a project needs more, teams can customise sources, memory, roles, checks, and autonomy.

01 · CONTEXT ASSEMBLY

Starts work with context

Receives the task, decisions, dependencies, and related artifacts before it runs—without the team having to explain the project again.

02 · TOOL LAYER

Acts instead of only answering

Creates and updates tasks, works with files, and saves results through approved Labs actions.

03 · STATE & CONTROL

Continues under human control

Retains progress between runs, respects permissions, and leaves a verifiable history of work.

Claude

Codex

Gemini

DeepSeek

Mistral

Local

OfferWind Bridge · agent runtime & work trace

Bridge records a verifiable history of agent work.

Bridge runs an AI agent in a local or cloud runner and links every action to a task: the context used, tool calls, file changes, human reviews, and the final result.

01
Persistent agent executionA task, status change, or schedule can start an agent without an open AI client. State persists between runs.
02
Observable human reviewBridge records proposed changes, rejected decisions, human edits, repeated iterations, and executed checks.
03
Evidence for a verified profileThe local trace joins GitHub events, CI results, and team confirmations while preserving the source and context of every signal.

Bridge helps build a proof-of-work profile from verified events. MCP can also control Labs from an AI client, but Bridge captures the full execution and review history.

Open-source release planned. The OWL Bridge public repository is not available yet.

offerwind-bridge — runner

Last login: today on ttys001

~ offerwind bridge start --project ow-research

09:42:10 bridge connecting to Labs...

09:42:10 connected wss://labs.offerwind.io

09:42:11 event task.assigned task_031

09:42:11 harness context assembled 12 artifacts

09:42:13 agent patch.proposed 6 files · sha 8f3a

09:48:06 human review.started maya

09:53:21 review change.requested auth check

10:07:14 test suite.completed 24 passed

10:08:02 trace evidence.sealed review verified

Model Context Protocol · client session

Control Labs from the AI client you already use.

MCP exposes OfferWind Labs tasks, files, and actions inside a compatible client. The person starts and controls the session, while the agent follows the same Harness permissions and project context.

01
No separate runnerConnect the MCP server to your client and call Labs tools from the current conversation.
02
Direct session controlYou decide when to invoke the agent, which context to provide, and which actions require approval.
03
The same Labs actionsBridge and MCP use the same tools, permissions, and project action log.

MCP is the alternative connection method. For event-driven work, background execution, and persistent state between sessions, Bridge is the better fit.

AI Client OfferWind MCP

AI Assistant ⌄ ow-research

Check the team’s task status and find blockers for the report.

✦

AI Assistant

I’ll first check the project context and the team’s active tasks.

⌘

Project context retrievedofferwind.get_project_context

✓

⌘

Team tasks checkedofferwind.list_team_tasks

✓

I found one blocker: Task 31 is waiting for Maya’s review. The remaining active tasks are on track.

Ask about the project

＋↑

Proof-of-work · verified with Bridge

A profile built from work,
not claims.

Bridge and OfferWind Labs integrations preserve a verifiable chain: who defined the task, what the agent proposed, how a person reviewed it, and what the team accepted. These events become a profile that can be shared with a recruiter or hiring manager.

PROJECT EVENT LOG RECORDING

PROJECT 04Candidate matching APITeam · 4 members · 3 agents

09:14MC
Added acceptance criteriaTask 18 · permissions & fallback behaviour
HUMAN
09:31✦
Proposed changesPatch 01 · 6 files · trace recorded
AGENT
10:02MC
Found an access-control errorReturned the patch to the agent with feedback
REVIEW
10:26M+✦
Corrected the behaviour and added a testPatch 02 · review link preserved
ITERATION
11:08✓
PR accepted, checks passed24 tests passed · teammate approved
ACCEPTED

BRIDGE EVIDENCE LAYER Links each action to a contributor, task, and outcome

01 Event
02 Artifact
03 Human review
04 Confirmation

EXAMPLES OF OBSERVABLE SIGNALS

Different work requires different evidence.

OfferWind does not judge a researcher by commit count or a developer by message volume. The profile presents signals relevant to each type of work.

01 · ENGINEERING

Engineering

AI-generated patches reviewed by a person
Errors found before merge
Added tests and CI results
Addressed PR feedback

02 · RESEARCH & DATA

Research and data

Verified sources and hypotheses
Findings linked to data and method
Reproducible analysis
Results accepted by the team

03 · PRODUCT & OPS

Product and operations

Tasks and acceptance criteria defined
Dependencies and blockers resolved
People and agents coordinated
Outcome shipped

04 · DESIGN & CREATIVE

Design and creative

Iterations and decision context
Team critique incorporated
Components and assets delivered
Final outcome accepted

FOR CANDIDATES

Show your role, not only the outcome.

The profile combines projects, artifacts, human review, and team confirmations. You decide which work episodes to share through a private link.

FOR COMPANIES

See evidence of work before the interview.

Recruiters can inspect the source of every signal and decide which actions and outcomes matter for a particular role.

Future project marketplace

Real projects that build experience, evidence, and income.

In the future, companies will be able to publish paid research and product challenges. Participants will choose a project and team, gain practical experience, build proof-of-work, and receive compensation for completed work.

Real company challenge
Payment for participation or an accepted outcome
Experience working in a human–AI team
Portable evidence of completed work

Join the early pilot ↗