verel 0.3.0

The agent framework where nothing is done until a grader returns a verdict — verification + grounded perception (AgentVision eyes), compounding memory, a fleet, agent-built tooling, and agent-run CI/CD.

Verel

The agent framework where nothing is "done" until a grader returns a verdict — checked by real senses including eyes (AgentVision) — and only verified work is allowed to compound into the fleet's shared memory.

Documents

• docs/VEREL_DESIGN.md — definitive architecture & build plan (positioning, the moat, the five organs, the Brain/memory architecture, the Fleet, the Verdict bus, AgentVision-as-eyes, claimable inventions, risks, phased roadmap, open decisions).
• docs/CRITIC_CONVERGENCE.md — the adversarial critic-loop score record that the design was iterated against until diminishing returns.

The five organs

Brain (memory)  ─┐
Fleet (agents managing agents) ─┤
Verdict bus (eval-driven everything) ─┼─► nothing merges on a self-asserted "done"
Senses (AgentVision eyes + logs/tests/metrics) ─┤
Tool-smith (agent-built tooling) ─┘

Install

pip install verel                 # core: the verdict bus + memory + fleet + tool-smith + CI
pip install "verel[sight]"        # + AgentVision (the eyes)
pip install "verel[mem0,mcp]"     # + rented memory backend, + MCP server

verel doctor                      # check the environment
verel heal --repo .               # self-healing CI: failing tests → agent fixes → green
verel loop dashboard.html         # fix a UI until AgentVision returns a pass verdict
verel fleet "fix the pages" --artifacts a.html b.html

Default LLM is Ollama Cloud (~/.config/ollama/key, model qwen3-coder:480b); set VEREL_LLM_PROVIDER=openai to switch. Claude is one branch away in agents/llm.py.

What's built (all five organs, end-to-end)

Organ	Module	What works
Verdict bus	verel.verdict	one schema for vision + tests + lint + types; advisory ceiling, grader attestation, scrubbed fingerprints, strict-subset stuck/progress
Eyes	verel.senses	AgentVision adapter — perception feeds the bus and memory
Agents	verel.agents	coder (fixes UIs) + code-fixer (patches source); Ollama Cloud
Brain	verel.memory	trust layer (LocalMemory / mem0), failure ledger + regression guard, consolidation, held-out attested promotion gate
Fleet	verel.fleet	single-writer scheduler (barriers/budget/WAL), LLM-driven manager, isolated git worktrees
Tool-smith	verel.toolsmith	detect → scaffold → test → register → reuse; signed registry; sandboxed exec
Agent-run CI/CD	verel.ci	tests/lint/type graders, inner-loop/pre-commit/pre-merge stages, self-healing, ci-medic, deterministic rollback engine, git hook + CLI
Surfaces	verel.cli, verel.mcp_server	verel CLI, MCP server, verel-ci

106 tests, 9 runnable demos. Code + module guide: src/verel/. Design & roadmap: docs/VEREL_DESIGN.md. Changelog: CHANGELOG.md.